From 7a68b81829cfe59ed132f2cee792d55977621b0a Mon Sep 17 00:00:00 2001 From: Faried Abu Zaid Date: Sat, 4 Jul 2026 14:32:11 +0200 Subject: [PATCH] Fix packaging layout for PyPI publication The wheel previously installed generic top-level packages `src` and `scripts`, forcing `import src.cjd_flows` and inviting collisions. Switch to a proper src-layout so the package installs as `cjd_flows`, stop shipping the experiment runner scripts, and rewrite all `src.cjd_flows` module references to `cjd_flows` (code, tests, and experiment configs). Remove the obsolete src/__init__.py. Verified: wheel contains only cjd_flows/, installs and imports in a fresh venv, full test suite passes against the editable install. Co-Authored-By: Claude Fable 5 --- experiments/cfair/config.yaml | 8 +-- experiments/cifar/cifar.yaml | 16 ++--- experiments/cifar/cifar_complementary.yaml | 16 ++--- experiments/cifar/cifar_jet.yaml | 12 ++-- experiments/fashion/config.yaml | 8 +-- experiments/fashion/fashion.yaml | 16 ++--- .../fashion/fashion_complementary.yaml | 16 ++--- experiments/fashion/fashionclasses_macow.yaml | 10 +-- .../fashionclasses_macow_complementary.yaml | 14 ++-- .../fashion/fashionclasses_veriflow.yaml | 16 ++--- ...fashionclasses_veriflow_complementary.yaml | 12 ++-- experiments/mnist/config_best.yaml | 8 +-- experiments/mnist/config_lu.yaml | 8 +-- experiments/mnist/mnist.yaml | 16 ++--- experiments/mnist/mnist_0_minimal.yaml | 8 +-- experiments/mnist/mnist_complementary.yaml | 16 ++--- experiments/mnist/mnist_digits.yaml | 16 ++--- .../mnist/mnist_digits_complementary.yaml | 12 ++-- .../mnist_digits_macow_complementary.yaml | 16 ++--- experiments/mnist/mnist_digits_minimal.yaml | 26 +++---- .../mnist_digits_minimal_radial_chi.yaml | 24 +++---- .../mnist_digits_minimal_radial_chi2.yaml | 16 ++--- ...ist_digits_minimal_radial_exponential.yaml | 12 ++-- .../mnist_digits_minimal_radial_logN.yaml | 12 ++-- .../mnist_digits_minimal_radial_weilbul.yaml | 26 +++---- .../mnist_digits_minimal_radialdists.yaml | 14 ++-- experiments/mnist/mnist_usflow.yaml | 10 +-- experiments/mnist/mnist_usflow_cpu.yaml | 12 ++-- .../mnist/mnist_usflow_cpu_gammamm.yaml | 16 ++--- experiments/mnist/mnist_usflow_gpu.yaml | 10 +-- experiments/mnist/mnist_usflow_hyperopt.yaml | 14 ++-- experiments/mnist/mnist_usflow_minimal.yaml | 32 ++++----- .../mnist/mnist_usflow_radial_mm_gpu.yaml | 14 ++-- experiments/synthetic/gaussian_mixture.yaml | 70 +++++++++---------- .../gaussian_mixture_standart_base.yaml | 34 ++++----- experiments/synthetic/synthetic.yaml | 10 +-- experiments/synthetic/synthetic_rad_logN.yaml | 12 ++-- pyproject.toml | 9 +-- scripts/eval.py | 6 +- scripts/run-experiment.py | 2 +- src/__init__.py | 0 src/cjd_flows/distributions.py | 6 +- src/cjd_flows/explib/eval.py | 2 +- src/cjd_flows/explib/hyperopt.py | 8 +-- src/cjd_flows/explib/visualization.py | 4 +- src/cjd_flows/flows.py | 6 +- src/cjd_flows/networks.py | 2 +- tests/cjd_flows/flows_test.py | 4 +- tests/cjd_flows/linalg_test.py | 2 +- tests/cjd_flows/networks_test.py | 4 +- tests/cjd_flows/transforms_test.py | 2 +- tests/explib/hyperopt_test.py | 2 +- tests/explib/mnist.yaml | 16 ++--- 53 files changed, 340 insertions(+), 343 deletions(-) delete mode 100644 src/__init__.py diff --git a/experiments/cfair/config.yaml b/experiments/cfair/config.yaml index 2f3a253..a3c9d3d 100644 --- a/experiments/cfair/config.yaml +++ b/experiments/cfair/config.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: cfair10_basedist_comparison experiments: - &exp_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_laplace scheduler: &scheduler __object__: ray.tune.schedulers.ASHAScheduler @@ -18,7 +18,7 @@ experiments: mode: min trial_config: dataset: &dataset - __object__: src.cjd_flows.experiments.datasets.CfairSplit + __object__: cjd_flows.experiments.datasets.CfairSplit digit: 0 epochs: &epochs 20000 patience: &patience 50 @@ -34,7 +34,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.NiceFlow + __class__: &model cjd_flows.flows.NiceFlow params: coupling_layers: &coupling_layers __eval__: tune.choice([ 2, 3, 4, 5]) diff --git a/experiments/cifar/cifar.yaml b/experiments/cifar/cifar.yaml index 9238583..5c4218e 100644 --- a/experiments/cifar/cifar.yaml +++ b/experiments/cifar/cifar.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_radial_logN skip: true device: cpu @@ -25,7 +25,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.Cifar10Split + __class__: cjd_flows.explib.datasets.Cifar10Split params: space_to_depth_factor: 4 dataloc: /home/faried/Projects/USFlows/data/cifar10 @@ -36,7 +36,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -44,7 +44,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -59,7 +59,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 48 c_hidden: @@ -78,14 +78,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: __eval__: float("1") loc: __eval__: torch.zeros([48, 8, 8]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.ones([1]).to("cpu") * 6 scale: diff --git a/experiments/cifar/cifar_complementary.yaml b/experiments/cifar/cifar_complementary.yaml index ef300fb..98ef16b 100644 --- a/experiments/cifar/cifar_complementary.yaml +++ b/experiments/cifar/cifar_complementary.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_radial_logN skip: true device: cuda @@ -25,7 +25,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.Cifar10Split + __class__: cjd_flows.explib.datasets.Cifar10Split params: space_to_depth_factor: 4 dataloc: /home/faried/Projects/USFlows/data/cifar10 @@ -36,7 +36,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -44,7 +44,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -59,7 +59,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 48 c_hidden: @@ -97,14 +97,14 @@ experiments: affine_conjugation: false base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cuda p: __eval__: float("1") loc: __eval__: torch.zeros([48, 8, 8]).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.ones([1]).to("cuda") * 6 scale: diff --git a/experiments/cifar/cifar_jet.yaml b/experiments/cifar/cifar_jet.yaml index 27ed868..128c045 100644 --- a/experiments/cifar/cifar_jet.yaml +++ b/experiments/cifar/cifar_jet.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: cifar_jet experiments: - &exp_jet - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cifar_jet_conditioner skip: false device: cuda @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.Cifar10Split + __class__: cjd_flows.explib.datasets.Cifar10Split params: space_to_depth_factor: 4 dataloc: /home/faried/Projects/USFlows/data/cifar10 @@ -35,7 +35,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-4 @@ -43,7 +43,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -62,7 +62,7 @@ experiments: # couplings, as in the Jet architecture. masktype: channel conditioner_cls: - __class__: src.cjd_flows.networks.JetConditioner + __class__: cjd_flows.networks.JetConditioner conditioner_args: in_dims: [48, 8, 8] # space-to-depth already patchified the image; keep 8x8=64 tokens diff --git a/experiments/fashion/config.yaml b/experiments/fashion/config.yaml index 25f6e52..cf6daa5 100644 --- a/experiments/fashion/config.yaml +++ b/experiments/fashion/config.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_basedist_comparison experiments: - &exp_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_laplace scheduler: &scheduler __object__: ray.tune.schedulers.ASHAScheduler @@ -18,7 +18,7 @@ experiments: mode: min trial_config: dataset: &dataset - __object__: src.cjd_flows.explib.datasets.FashionMnistSplit + __object__: cjd_flows.explib.datasets.FashionMnistSplit label: 0 epochs: &epochs 10000 patience: &patience 500 @@ -34,7 +34,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.NiceFlow + __class__: &model cjd_flows.flows.NiceFlow params: coupling_layers: &coupling_layers __eval__: tune.choice([2, 3, 4, 6, 8]) diff --git a/experiments/fashion/fashion.yaml b/experiments/fashion/fashion.yaml index f816022..8684247 100644 --- a/experiments/fashion/fashion.yaml +++ b/experiments/fashion/fashion.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_radial_logN skip: true device: cpu @@ -25,7 +25,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.FashionMnistSplit + __class__: cjd_flows.explib.datasets.FashionMnistSplit params: space_to_depth_factor: 4 dataloc: /home/faried/Projects/USFlows/data/fashion @@ -36,7 +36,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -44,7 +44,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -59,7 +59,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 48 c_hidden: @@ -78,14 +78,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: __eval__: float("1") loc: __eval__: torch.zeros([48, 8, 8]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.ones([1]).to("cpu") * 6 scale: diff --git a/experiments/fashion/fashion_complementary.yaml b/experiments/fashion/fashion_complementary.yaml index 9629f5d..c2bd4ce 100644 --- a/experiments/fashion/fashion_complementary.yaml +++ b/experiments/fashion/fashion_complementary.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_radial_logN skip: true device: cuda @@ -25,7 +25,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.FashionMnistSplit + __class__: cjd_flows.explib.datasets.FashionMnistSplit params: space_to_depth_factor: 4 dataloc: /home/faried/Projects/USFlows/data/fashion @@ -36,7 +36,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -44,7 +44,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -59,7 +59,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 48 c_hidden: @@ -98,14 +98,14 @@ experiments: affine_conjugation: false base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cuda p: __eval__: float("1") loc: __eval__: torch.zeros([48, 8, 8]).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.ones([1]).to("cuda") * 6 scale: diff --git a/experiments/fashion/fashionclasses_macow.yaml b/experiments/fashion/fashionclasses_macow.yaml index 1fe5ef7..9ff381a 100644 --- a/experiments/fashion/fashionclasses_macow.yaml +++ b/experiments/fashion/fashionclasses_macow.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation_macow experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_normal skip: false device: cpu @@ -35,7 +35,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -43,7 +43,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -58,7 +58,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: diff --git a/experiments/fashion/fashionclasses_macow_complementary.yaml b/experiments/fashion/fashionclasses_macow_complementary.yaml index 7d9a1fc..348855b 100644 --- a/experiments/fashion/fashionclasses_macow_complementary.yaml +++ b/experiments/fashion/fashionclasses_macow_complementary.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation_macow experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_normal skip: false device: cuda @@ -35,7 +35,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -43,7 +43,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -58,7 +58,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -77,14 +77,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cuda p: __eval__: float("1") loc: __eval__: torch.zeros([16, 7, 7]).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cuda") * 75 rate: diff --git a/experiments/fashion/fashionclasses_veriflow.yaml b/experiments/fashion/fashionclasses_veriflow.yaml index 2c0edd4..38f129d 100644 --- a/experiments/fashion/fashionclasses_veriflow.yaml +++ b/experiments/fashion/fashionclasses_veriflow.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation_veriflow experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_radial_logN skip: false device: cpu @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.FashionMnistSplit + __class__: cjd_flows.explib.datasets.FashionMnistSplit params: space_to_depth_factor: 4 dataloc: /home/faried/Projects/USFlows/data/fashion @@ -35,7 +35,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -43,7 +43,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -58,7 +58,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -77,14 +77,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: __eval__: float("1") loc: __eval__: torch.zeros([16, 7, 7]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * 75 rate: diff --git a/experiments/fashion/fashionclasses_veriflow_complementary.yaml b/experiments/fashion/fashionclasses_veriflow_complementary.yaml index bb8c64b..96e3fa4 100644 --- a/experiments/fashion/fashionclasses_veriflow_complementary.yaml +++ b/experiments/fashion/fashionclasses_veriflow_complementary.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: fashion_ablation_veriflow experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: cfair_full_radial_logN skip: false device: cuda @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.FashionMnistSplit + __class__: cjd_flows.explib.datasets.FashionMnistSplit params: space_to_depth_factor: 4 dataloc: /home/faried/Projects/USFlows/data/fashion @@ -35,7 +35,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -43,7 +43,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -58,7 +58,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: diff --git a/experiments/mnist/config_best.yaml b/experiments/mnist/config_best.yaml index 1fee4ce..4f2ed8e 100644 --- a/experiments/mnist/config_best.yaml +++ b/experiments/mnist/config_best.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_basedist_comparison experiments: - &exp_laplace_best - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_normal_best scheduler: &scheduler __object__: ray.tune.schedulers.ASHAScheduler @@ -18,7 +18,7 @@ experiments: mode: min trial_config: dataset: &dataset - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 epochs: &epochs 10000 patience: &patience 50 @@ -32,7 +32,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.NiceFlow + __class__: &model cjd_flows.flows.NiceFlow params: coupling_layers: 4 coupling_nn_layers: [50, 50] diff --git a/experiments/mnist/config_lu.yaml b/experiments/mnist/config_lu.yaml index 034e97e..8d8b649 100644 --- a/experiments/mnist/config_lu.yaml +++ b/experiments/mnist/config_lu.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.baseExperimentCollection +__object__: cjd_flows.explib.baseExperimentCollection name: mnist_basedist_comparison experiments: - &exp_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_laplace scheduler: &scheduler __object__: ray.tune.schedulers.ASHAScheduler @@ -18,7 +18,7 @@ experiments: mode: min trial_config: dataset: &dataset - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 epochs: &epochs 20000 patience: &patience 50 @@ -34,7 +34,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.LUFlow + __class__: &model cjd_flows.flows.LUFlow params: n_layers: &n_layers __eval__: tune.choice([ 2, 3, 4, 5, 6, 7, 8, 9, 10]) diff --git a/experiments/mnist/mnist.yaml b/experiments/mnist/mnist.yaml index cec2734..90ee9f4 100644 --- a/experiments/mnist/mnist.yaml +++ b/experiments/mnist/mnist.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_full_radial_logN device: cpu skip: False @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.MnistSplit + __class__: cjd_flows.explib.datasets.MnistSplit params: dataloc: /home/faried/Projects/USFlows/data/mnist space_to_depth_factor: 4 @@ -35,7 +35,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -43,7 +43,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -58,7 +58,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -77,14 +77,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: __eval__: float("1") loc: __eval__: torch.zeros([16, 7, 7]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.ones([1]).to("cpu") * 6 scale: diff --git a/experiments/mnist/mnist_0_minimal.yaml b/experiments/mnist/mnist_0_minimal.yaml index 06e64d3..5c8b6d5 100644 --- a/experiments/mnist/mnist_0_minimal.yaml +++ b/experiments/mnist/mnist_0_minimal.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_nice_lu_laplace scheduler: &scheduler __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: &dataset - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: &epochs 200000 @@ -38,7 +38,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.NiceFlow + __class__: &model cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false diff --git a/experiments/mnist/mnist_complementary.yaml b/experiments/mnist/mnist_complementary.yaml index 99d8145..38fd5fc 100644 --- a/experiments/mnist/mnist_complementary.yaml +++ b/experiments/mnist/mnist_complementary.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_full_radial_logN device: cuda skip: False @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.MnistSplit + __class__: cjd_flows.explib.datasets.MnistSplit params: dataloc: /home/faried/Projects/USFlows/data/mnist space_to_depth_factor: 4 @@ -35,7 +35,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 @@ -43,7 +43,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -58,7 +58,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -97,14 +97,14 @@ experiments: affine_conjugation: false base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cuda p: __eval__: float("1") loc: __eval__: torch.zeros([16, 7, 7]).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.ones([1]).to("cuda") * 6 scale: diff --git a/experiments/mnist/mnist_digits.yaml b/experiments/mnist/mnist_digits.yaml index 028473a..4c54f30 100644 --- a/experiments/mnist/mnist_digits.yaml +++ b/experiments/mnist/mnist_digits.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_gigits_logN experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist0 device: cpu skip: false @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.MnistSplit + __class__: cjd_flows.explib.datasets.MnistSplit params: dataloc: /home/faried/Projects/USFlows/data/mnist space_to_depth_factor: 4 @@ -36,7 +36,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-4 @@ -44,7 +44,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -59,7 +59,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -78,14 +78,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: __eval__: float("1") loc: __eval__: torch.zeros([16, 7, 7]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * 75 rate: diff --git a/experiments/mnist/mnist_digits_complementary.yaml b/experiments/mnist/mnist_digits_complementary.yaml index 47ca33f..5745267 100644 --- a/experiments/mnist/mnist_digits_complementary.yaml +++ b/experiments/mnist/mnist_digits_complementary.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_gigits_logN experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist0 device: cuda skip: false @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.MnistSplit + __class__: cjd_flows.explib.datasets.MnistSplit params: dataloc: /home/faried/Projects/USFlows/data/mnist space_to_depth_factor: 4 @@ -36,7 +36,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-4 @@ -44,7 +44,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -59,7 +59,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: diff --git a/experiments/mnist/mnist_digits_macow_complementary.yaml b/experiments/mnist/mnist_digits_macow_complementary.yaml index f16030d..86ff1c9 100644 --- a/experiments/mnist/mnist_digits_macow_complementary.yaml +++ b/experiments/mnist/mnist_digits_macow_complementary.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_gigits_logN experiments: - &exp_rad_logN - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist0 device: cuda skip: false @@ -24,7 +24,7 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.MnistSplit + __class__: cjd_flows.explib.datasets.MnistSplit params: dataloc: /home/faried/Projects/USFlows/data/mnist space_to_depth_factor: 4 @@ -36,7 +36,7 @@ experiments: __eval__: tune.choice([32]) optim_cfg: optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-4 @@ -44,7 +44,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -59,7 +59,7 @@ experiments: lu_transform: 0 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -78,14 +78,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cuda p: __eval__: float("1") loc: __eval__: torch.zeros([16, 7, 7]).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cuda") * 75 rate: diff --git a/experiments/mnist/mnist_digits_minimal.yaml b/experiments/mnist/mnist_digits_minimal.yaml index f128980..efb82ef 100644 --- a/experiments/mnist/mnist_digits_minimal.yaml +++ b/experiments/mnist/mnist_digits_minimal.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_digit_scaled_small_models experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_0 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: 200000 @@ -37,7 +37,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.NiceFlow + __class__: cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false @@ -65,62 +65,62 @@ experiments: name: mnist_1 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 1 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_2 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 2 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_3 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 3 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_4 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 4 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_5 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 5 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_6 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 6 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_7 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 7 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_8 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 8 scale: true - __overwrites__: *exp_nice_lu_laplace name: mnist_9 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 9 scale: true \ No newline at end of file diff --git a/experiments/mnist/mnist_digits_minimal_radial_chi.yaml b/experiments/mnist/mnist_digits_minimal_radial_chi.yaml index 2101fd2..171ac36 100644 --- a/experiments/mnist/mnist_digits_minimal_radial_chi.yaml +++ b/experiments/mnist/mnist_digits_minimal_radial_chi.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_digit_scaled_small_models_l1rad experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_0_chi_2.0 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: 200000 @@ -37,7 +37,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.NiceFlow + __class__: cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false @@ -55,12 +55,12 @@ experiments: split_dim: __eval__: tune.choice([10 + i for i in range(41)]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") radial_distribution: - __object__: src.cjd_flows.distributions.Chi + __object__: cjd_flows.distributions.Chi df: __eval__: torch.Tensor([2.0]) use_lu: true @@ -71,12 +71,12 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") radial_distribution: - __object__: src.cjd_flows.distributions.Chi + __object__: cjd_flows.distributions.Chi df: __eval__: torch.Tensor([1.0]) - __overwrites__: *exp_nice_lu_laplace @@ -86,12 +86,12 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") radial_distribution: - __object__: src.cjd_flows.distributions.Chi + __object__: cjd_flows.distributions.Chi df: __eval__: torch.Tensor([3.0]) - __overwrites__: *exp_nice_lu_laplace @@ -101,11 +101,11 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") radial_distribution: - __object__: src.cjd_flows.distributions.Chi + __object__: cjd_flows.distributions.Chi df: __eval__: torch.Tensor([4.0]) diff --git a/experiments/mnist/mnist_digits_minimal_radial_chi2.yaml b/experiments/mnist/mnist_digits_minimal_radial_chi2.yaml index 55ba4bf..6f48c9f 100644 --- a/experiments/mnist/mnist_digits_minimal_radial_chi2.yaml +++ b/experiments/mnist/mnist_digits_minimal_radial_chi2.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_digit_scaled_small_models_l1rad experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_0_chi2_1 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: 200000 @@ -37,7 +37,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.NiceFlow + __class__: cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false @@ -55,7 +55,7 @@ experiments: split_dim: __eval__: tune.choice([10 + i for i in range(41)]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -70,7 +70,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -84,7 +84,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -98,7 +98,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") diff --git a/experiments/mnist/mnist_digits_minimal_radial_exponential.yaml b/experiments/mnist/mnist_digits_minimal_radial_exponential.yaml index 435e257..bf6ad46 100644 --- a/experiments/mnist/mnist_digits_minimal_radial_exponential.yaml +++ b/experiments/mnist/mnist_digits_minimal_radial_exponential.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_digit_scaled_small_models_l1rad experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_0_logN_0.75 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: 200000 @@ -37,7 +37,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.NiceFlow + __class__: cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false @@ -55,7 +55,7 @@ experiments: split_dim: __eval__: tune.choice([10 + i for i in range(41)]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -71,7 +71,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") diff --git a/experiments/mnist/mnist_digits_minimal_radial_logN.yaml b/experiments/mnist/mnist_digits_minimal_radial_logN.yaml index 7eda5bb..3c3829e 100644 --- a/experiments/mnist/mnist_digits_minimal_radial_logN.yaml +++ b/experiments/mnist/mnist_digits_minimal_radial_logN.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_digit_scaled_small_models_l1rad experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_0_logN_0.75 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: 200000 @@ -37,7 +37,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.NiceFlow + __class__: cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false @@ -55,7 +55,7 @@ experiments: split_dim: __eval__: tune.choice([10 + i for i in range(41)]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -73,7 +73,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") diff --git a/experiments/mnist/mnist_digits_minimal_radial_weilbul.yaml b/experiments/mnist/mnist_digits_minimal_radial_weilbul.yaml index 48d2e3b..0a2ec9c 100644 --- a/experiments/mnist/mnist_digits_minimal_radial_weilbul.yaml +++ b/experiments/mnist/mnist_digits_minimal_radial_weilbul.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_digit_scaled_small_models_l1rad experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_0_weilbull_1.0 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: 200000 @@ -37,7 +37,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.NiceFlow + __class__: cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false @@ -55,7 +55,7 @@ experiments: split_dim: __eval__: tune.choice([10 + i for i in range(41)]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -72,7 +72,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -88,7 +88,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -104,7 +104,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -120,7 +120,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -136,7 +136,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -152,7 +152,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -168,7 +168,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -184,7 +184,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") diff --git a/experiments/mnist/mnist_digits_minimal_radialdists.yaml b/experiments/mnist/mnist_digits_minimal_radialdists.yaml index 75a8dbd..faef64d 100644 --- a/experiments/mnist/mnist_digits_minimal_radialdists.yaml +++ b/experiments/mnist/mnist_digits_minimal_radialdists.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_digit_scaled_small_models_l1rad experiments: - &exp_nice_lu_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_0_logN scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -21,7 +21,7 @@ experiments: images: true "image_shape": [10, 10] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit digit: 0 scale: true epochs: 200000 @@ -37,7 +37,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.NiceFlow + __class__: cjd_flows.flows.NiceFlow params: masktype: alternate soft_training: false @@ -55,7 +55,7 @@ experiments: split_dim: __eval__: tune.choice([10 + i for i in range(41)]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -73,7 +73,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") @@ -87,7 +87,7 @@ experiments: params: base_distribution: __exact__: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: 1.0 loc: __eval__: torch.zeros(100).to("cpu") diff --git a/experiments/mnist/mnist_usflow.yaml b/experiments/mnist/mnist_usflow.yaml index ded8c0c..c6d9793 100644 --- a/experiments/mnist/mnist_usflow.yaml +++ b/experiments/mnist/mnist_usflow.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_full_radial_logN scheduler: &scheduler __object__: ray.tune.schedulers.ASHAScheduler @@ -22,7 +22,7 @@ experiments: images: false "image_shape": [28, 28] dataset: &dataset - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 2 device: cuda epochs: &epochs 200000 @@ -39,7 +39,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.USFlow + __class__: &model cjd_flows.flows.USFlow params: soft_training: false training_noise_prior: @@ -50,7 +50,7 @@ experiments: prior_scale: 1.0 coupling_blocks: 1 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 4 num_layers: 3 diff --git a/experiments/mnist/mnist_usflow_cpu.yaml b/experiments/mnist/mnist_usflow_cpu.yaml index 0959042..cf67da7 100644 --- a/experiments/mnist/mnist_usflow_cpu.yaml +++ b/experiments/mnist/mnist_usflow_cpu.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_laplace0 - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist0 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -22,7 +22,7 @@ experiments: images: false image_shape: [28, 28] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cpu digit: 0 @@ -40,7 +40,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: true training_noise_prior: @@ -53,7 +53,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.CondConvNet2D + __class__: cjd_flows.networks.CondConvNet2D conditioner_args: c_in: 16 c_hidden: 32 @@ -66,7 +66,7 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.Laplace + __object__: cjd_flows.distributions.Laplace loc: __eval__: torch.zeros([16, 7, 7]).to("cpu") scale: diff --git a/experiments/mnist/mnist_usflow_cpu_gammamm.yaml b/experiments/mnist/mnist_usflow_cpu_gammamm.yaml index 0a4d8c1..0312bf8 100644 --- a/experiments/mnist/mnist_usflow_cpu_gammamm.yaml +++ b/experiments/mnist/mnist_usflow_cpu_gammamm.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_laplace0 - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist0 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -25,7 +25,7 @@ experiments: image_shape: [28, 28] dataset: class: - __class__: src.cjd_flows.explib.datasets.MnistSplit + __class__: cjd_flows.explib.datasets.MnistSplit params: dataloc: /Users/fariedabuzaid/Projects/veriflow/data space_to_depth_factor: 4 @@ -35,14 +35,14 @@ experiments: __eval__: tune.choice([32]) optim_cfg: optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-4 weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: false training_noise_prior: @@ -55,7 +55,7 @@ experiments: lu_transform: 1 householder: 0 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: 32 @@ -73,7 +73,7 @@ experiments: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: class: - __class__: src.cjd_flows.distributions.RadialMM + __class__: cjd_flows.distributions.RadialMM params: device: cpu p: 2.0 @@ -81,7 +81,7 @@ experiments: __eval__: torch.randn([20, 16, 7, 7]).to("cpu") * 0.1 norm_distribution: class: - __class__: src.cjd_flows.distributions.LogNormal + __class__: cjd_flows.distributions.LogNormal params: loc: __eval__: torch.ones([20, 1, 1, 1]).to("cpu") * 1 diff --git a/experiments/mnist/mnist_usflow_gpu.yaml b/experiments/mnist/mnist_usflow_gpu.yaml index 78a2f32..fb80372 100644 --- a/experiments/mnist/mnist_usflow_gpu.yaml +++ b/experiments/mnist/mnist_usflow_gpu.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_full_laplace scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -22,7 +22,7 @@ experiments: images: false image_shape: [28, 28] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 2 device: cuda # Options: [cpu, cuda] epochs: 200000 @@ -39,7 +39,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: false training_noise_prior: @@ -52,7 +52,7 @@ experiments: lu_transform: 1 householder: 1 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 4 num_layers: 4 diff --git a/experiments/mnist/mnist_usflow_hyperopt.yaml b/experiments/mnist/mnist_usflow_hyperopt.yaml index 24cf42a..b0b87d2 100644 --- a/experiments/mnist/mnist_usflow_hyperopt.yaml +++ b/experiments/mnist/mnist_usflow_hyperopt.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_full_laplace scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -22,7 +22,7 @@ experiments: images: false image_shape: [28, 28] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 digit: 3 device: cuda @@ -39,7 +39,7 @@ experiments: weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: true in_dims: [16, 7, 7] @@ -59,7 +59,7 @@ experiments: affine_conjugation: __eval__: tune.choice([True, False]) conditioner_cls: - __class__: src.cjd_flows.networks.CondConvNet2D + __class__: cjd_flows.networks.CondConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -80,13 +80,13 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cuda p: 1.0 loc: __eval__: torch.zeros(16, 7, 7).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.zeros(1).to("cuda") scale: diff --git a/experiments/mnist/mnist_usflow_minimal.yaml b/experiments/mnist/mnist_usflow_minimal.yaml index 4d3e86a..c5c8f8c 100644 --- a/experiments/mnist/mnist_usflow_minimal.yaml +++ b/experiments/mnist/mnist_usflow_minimal.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_laplace0 - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist0 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -22,7 +22,7 @@ experiments: images: false image_shape: [28, 28] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 0 @@ -40,7 +40,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: true training_noise_prior: @@ -53,7 +53,7 @@ experiments: lu_transform: 1 householder: 1 conditioner_cls: - __class__: src.cjd_flows.networks.CondConvNet2D + __class__: cjd_flows.networks.CondConvNet2D conditioner_args: c_in: 16 c_hidden: 32 @@ -67,13 +67,13 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cuda p: 1.0 loc: __eval__: torch.zeros([16, 7, 7]).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.ones([1]).to("cuda") * 4.5 scale: @@ -84,7 +84,7 @@ experiments: name: mnist1 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 1 @@ -93,7 +93,7 @@ experiments: name: mnist2 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 2 @@ -102,7 +102,7 @@ experiments: name: mnist3 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 3 @@ -111,7 +111,7 @@ experiments: name: mnist4 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 4 @@ -120,7 +120,7 @@ experiments: name: mnist5 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 5 @@ -129,7 +129,7 @@ experiments: name: mnist6 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 6 @@ -138,7 +138,7 @@ experiments: name: mnist7 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 7 @@ -147,7 +147,7 @@ experiments: name: mnist8 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 8 @@ -156,7 +156,7 @@ experiments: name: mnist9 trial_config: dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda digit: 9 diff --git a/experiments/mnist/mnist_usflow_radial_mm_gpu.yaml b/experiments/mnist/mnist_usflow_radial_mm_gpu.yaml index 64d7ace..94d1932 100644 --- a/experiments/mnist/mnist_usflow_radial_mm_gpu.yaml +++ b/experiments/mnist/mnist_usflow_radial_mm_gpu.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation experiments: - &exp_laplace - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist_full_laplace scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -22,7 +22,7 @@ experiments: images: false image_shape: [28, 28] dataset: - __object__: src.cjd_flows.explib.datasets.MnistSplit + __object__: cjd_flows.explib.datasets.MnistSplit space_to_depth_factor: 4 device: cuda # Options: [cpu, cuda] digit: 0 @@ -40,7 +40,7 @@ experiments: model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: true training_noise_prior: @@ -53,7 +53,7 @@ experiments: lu_transform: 1 householder: 1 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: 64 @@ -66,11 +66,11 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialMM + __object__: cjd_flows.distributions.RadialMM loc: __eval__: torch.randn([10, 16, 7, 7]).to("cuda") norm_distribution: - __object__: src.cjd_flows.distributions.LogNormal + __object__: cjd_flows.distributions.LogNormal loc: __eval__: torch.zeros([10, 1, 1, 1]).to("cuda") scale: diff --git a/experiments/synthetic/gaussian_mixture.yaml b/experiments/synthetic/gaussian_mixture.yaml index a15f57e..5da3436 100644 --- a/experiments/synthetic/gaussian_mixture.yaml +++ b/experiments/synthetic/gaussian_mixture.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: gaussian_mixture_experiments experiments: - &exp2d - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: gaussian_mixture_2D device: cpu scheduler: &scheduler @@ -23,10 +23,10 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.DistributionSplit + __class__: cjd_flows.explib.datasets.DistributionSplit params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0], [1.0, 1.0]]) covariance_matrix: @@ -42,14 +42,14 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -74,14 +74,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: __eval__: float("1") loc: __eval__: torch.zeros([2]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([2.0])) rate: @@ -95,7 +95,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0], [1.0, 1.0, 1.0]]) covariance_matrix: @@ -110,13 +110,13 @@ experiments: hidden_dims: [32, 32] param_dims: [3] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([3]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([3.0])) rate: @@ -130,7 +130,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -145,13 +145,13 @@ experiments: hidden_dims: [32, 32] param_dims: [4] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([4]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([4.0])) rate: @@ -165,7 +165,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -180,13 +180,13 @@ experiments: hidden_dims: [32, 32] param_dims: [5] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([5]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([5.0])) rate: @@ -200,7 +200,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -215,13 +215,13 @@ experiments: hidden_dims: [32, 32] param_dims: [6] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([6]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([6.0])) rate: @@ -235,7 +235,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -250,13 +250,13 @@ experiments: hidden_dims: [32, 32] param_dims: [7] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([7]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([7.0])) rate: @@ -270,7 +270,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -285,13 +285,13 @@ experiments: hidden_dims: [32, 32] param_dims: [8] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([8]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([8.0])) rate: @@ -305,7 +305,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -320,13 +320,13 @@ experiments: hidden_dims: [32, 32] param_dims: [9] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([9]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([9.0])) rate: @@ -340,7 +340,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -355,13 +355,13 @@ experiments: hidden_dims: [32, 32] param_dims: [10] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([10]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([10.0])) rate: @@ -375,7 +375,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0] * 100, [1.0] * 100]) covariance_matrix: @@ -390,13 +390,13 @@ experiments: hidden_dims: [32, 32] param_dims: [100] base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution p: __eval__: float("1") loc: __eval__: torch.zeros([100]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([20]).to("cpu") * torch.sqrt(torch.tensor([100.0])) rate: diff --git a/experiments/synthetic/gaussian_mixture_standart_base.yaml b/experiments/synthetic/gaussian_mixture_standart_base.yaml index aa66eea..d03743b 100644 --- a/experiments/synthetic/gaussian_mixture_standart_base.yaml +++ b/experiments/synthetic/gaussian_mixture_standart_base.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: gaussian_mixture_experiments experiments: - &exp2d - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: gaussian_mixture_2D device: cpu scheduler: &scheduler @@ -23,10 +23,10 @@ experiments: "image_shape": [28, 28] dataset: &dataset class: - __class__: src.cjd_flows.explib.datasets.DistributionSplit + __class__: cjd_flows.explib.datasets.DistributionSplit params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0], [1.0, 1.0]]) covariance_matrix: @@ -42,14 +42,14 @@ experiments: __eval__: tune.choice([32]) optim_cfg: &optim optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -74,7 +74,7 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.Normal + __object__: cjd_flows.distributions.Normal loc: __eval__: torch.zeros([2]).to("cpu") scale: @@ -86,7 +86,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0], [1.0, 1.0, 1.0]]) covariance_matrix: @@ -101,7 +101,7 @@ experiments: hidden_dims: [32, 32] param_dims: [3] base_distribution: - __object__: src.cjd_flows.distributions.Normal + __object__: cjd_flows.distributions.Normal loc: __eval__: torch.zeros([3]).to("cpu") - __overwrites__: *exp2d @@ -110,7 +110,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -133,7 +133,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -156,7 +156,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -179,7 +179,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -202,7 +202,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -225,7 +225,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -248,7 +248,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]]) covariance_matrix: @@ -271,7 +271,7 @@ experiments: dataset: params: distribution: - __object__: src.cjd_flows.distributions.GMM + __object__: cjd_flows.distributions.GMM loc: __eval__: torch.tensor([[-1.0] * 100, [1.0] * 100]) covariance_matrix: diff --git a/experiments/synthetic/synthetic.yaml b/experiments/synthetic/synthetic.yaml index e16c7e6..5ca5362 100644 --- a/experiments/synthetic/synthetic.yaml +++ b/experiments/synthetic/synthetic.yaml @@ -1,14 +1,14 @@ --- # Runs a series of experiments on various synthetic 2D datasets -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: synthetic_basedist_comparison experiments: - &main - __object__: src.cjd_flows.explib.base.ExperimentCollection + __object__: cjd_flows.explib.base.ExperimentCollection name: blobs experiments: - &main_normal - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: normal scheduler: &scheduler __object__: ray.tune.schedulers.ASHAScheduler @@ -26,7 +26,7 @@ experiments: images: false scatter: true dataset: &dataset - __object__: src.cjd_flows.explib.datasets.SyntheticSplit + __object__: cjd_flows.explib.datasets.SyntheticSplit generator: circles params_train: ¶ms_train n_samples: 100000 @@ -48,7 +48,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.NiceFlow + __class__: &model cjd_flows.flows.NiceFlow params: soft_training: true training_noise_prior: diff --git a/experiments/synthetic/synthetic_rad_logN.yaml b/experiments/synthetic/synthetic_rad_logN.yaml index 402ad9a..d885207 100644 --- a/experiments/synthetic/synthetic_rad_logN.yaml +++ b/experiments/synthetic/synthetic_rad_logN.yaml @@ -1,14 +1,14 @@ --- # Runs a series of experiments on various synthetic 2D datasets -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: synthetic_basedist_comparison experiments: - &main - __object__: src.cjd_flows.explib.base.ExperimentCollection + __object__: cjd_flows.explib.base.ExperimentCollection name: blobs experiments: - &main_normal - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: normal device: cpu skip: true @@ -28,7 +28,7 @@ experiments: images: false scatter: true dataset: &dataset - __object__: src.cjd_flows.explib.datasets.SyntheticSplit + __object__: cjd_flows.explib.datasets.SyntheticSplit generator: circles params_train: ¶ms_train n_samples: 100000 @@ -50,7 +50,7 @@ experiments: model_cfg: type: - __class__: &model src.cjd_flows.flows.NiceFlow + __class__: &model cjd_flows.flows.NiceFlow params: soft_training: true training_noise_prior: @@ -67,7 +67,7 @@ experiments: __eval__: tune.choice([torch.nn.ReLU()]) split_dim: &split_dim 1 base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: 1.0 loc: diff --git a/pyproject.toml b/pyproject.toml index 1434448..5e1c780 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -60,12 +60,9 @@ Repository = "https://github.com/USFlows/cjd-flows" Documentation = "https://github.com/USFlows/cjd-flows/tree/main/docs" [tool.poetry] -packages = [{include = "src/cjd_flows"}, {include = "scripts"}] - -# -#[tool.setuptools.packages.find] -#where = ["src"] -# +# src-layout: installs as the top-level `cjd_flows` package. +# The experiment runner scripts are repo tooling and are not shipped. +packages = [{include = "cjd_flows", from = "src"}] [tool.poetry.group.dev] diff --git a/scripts/eval.py b/scripts/eval.py index 0662c7b..7dc2b7a 100644 --- a/scripts/eval.py +++ b/scripts/eval.py @@ -11,8 +11,8 @@ from sklearn.neighbors import KernelDensity import pandas as pd -from src.cjd_flows.explib.config_parser import from_checkpoint -from src.cjd_flows.distributions import Independent +from cjd_flows.explib.config_parser import from_checkpoint +from cjd_flows.distributions import Independent import os import torch import numpy as np @@ -567,7 +567,7 @@ def _pp_plot_single(evaluator, ax, n_samples, label=None, color=None): model = from_checkpoint(pkl_path, pt_path) # Load test set - from src.cjd_flows.explib.datasets import MnistDequantized + from cjd_flows.explib.datasets import MnistDequantized mnisti = MnistDequantized(dataloc="/home/faried/Projects/USFlows/data/mnist", space_to_depth_factor=4, digit=i, train=False)[:1000][0] evaluator = RadialFlowEvaluator(model, mnisti) diff --git a/scripts/run-experiment.py b/scripts/run-experiment.py index 1b1e1b4..df9ea83 100755 --- a/scripts/run-experiment.py +++ b/scripts/run-experiment.py @@ -3,7 +3,7 @@ import click -from src.cjd_flows.explib.config_parser import read_config +from cjd_flows.explib.config_parser import read_config Pathable = T.Union[str, os.PathLike] # In principle one can cast it to os.path.Path import torch diff --git a/src/__init__.py b/src/__init__.py deleted file mode 100644 index e69de29..0000000 diff --git a/src/cjd_flows/distributions.py b/src/cjd_flows/distributions.py index 2e94b9d..13b7946 100644 --- a/src/cjd_flows/distributions.py +++ b/src/cjd_flows/distributions.py @@ -1,7 +1,7 @@ from typing import Dict, Iterable, Optional, Union -from src.cjd_flows.transforms import Rotation, CompositeRotation -from src.cjd_flows.linalg import random_orthonormal_matrix -from src.cjd_flows.utils import inv_softplus +from cjd_flows.transforms import Rotation, CompositeRotation +from cjd_flows.linalg import random_orthonormal_matrix +from cjd_flows.utils import inv_softplus import pyro import math diff --git a/src/cjd_flows/explib/eval.py b/src/cjd_flows/explib/eval.py index 634037d..c8e24d3 100644 --- a/src/cjd_flows/explib/eval.py +++ b/src/cjd_flows/explib/eval.py @@ -11,7 +11,7 @@ from scipy.stats import binomtest, wilcoxon from sklearn.neighbors import KernelDensity -from src.cjd_flows.distributions import RadialDistribution +from cjd_flows.distributions import RadialDistribution class RadialFlowEvaluator: def __init__(self, flow, data, device='cpu', p: Optional[float] = None, norm_distribution: Optional[torch.distributions.Distribution] = None, loc: Optional[torch.Tensor] = None): diff --git a/src/cjd_flows/explib/hyperopt.py b/src/cjd_flows/explib/hyperopt.py index cc508f7..0f56910 100644 --- a/src/cjd_flows/explib/hyperopt.py +++ b/src/cjd_flows/explib/hyperopt.py @@ -19,10 +19,10 @@ from ray import tune from ray.air import RunConfig, session -from src.cjd_flows.explib.base import Experiment -from src.cjd_flows.explib.config_parser import from_checkpoint, create_objects_from_classes -from src.cjd_flows.networks import AdditiveAffineNN -from src.cjd_flows.transforms import ScaleTransform +from cjd_flows.explib.base import Experiment +from cjd_flows.explib.config_parser import from_checkpoint, create_objects_from_classes +from cjd_flows.networks import AdditiveAffineNN +from cjd_flows.transforms import ScaleTransform diff --git a/src/cjd_flows/explib/visualization.py b/src/cjd_flows/explib/visualization.py index 241a36a..96a5f16 100644 --- a/src/cjd_flows/explib/visualization.py +++ b/src/cjd_flows/explib/visualization.py @@ -1,10 +1,10 @@ from typing import Dict, Iterable, Literal from matplotlib import pyplot as plt import numpy as np -from src.cjd_flows.explib import datasets +from cjd_flows.explib import datasets import torch -from src.cjd_flows.flows import Flow +from cjd_flows.flows import Flow Norm = Literal[-1, 1, 2] SampleType = Literal["conditional", "boundary", "boundary_basis"] diff --git a/src/cjd_flows/flows.py b/src/cjd_flows/flows.py index 85334ba..b48014d 100644 --- a/src/cjd_flows/flows.py +++ b/src/cjd_flows/flows.py @@ -5,10 +5,10 @@ from pyro import distributions as dist from typing import List, Dict, Literal, Any, Iterable, Optional, Type, Union, Tuple import torch -from src.cjd_flows.distributions import RadialDistribution, Independent -from src.cjd_flows.sophia import SophiaG +from cjd_flows.distributions import RadialDistribution, Independent +from cjd_flows.sophia import SophiaG -from src.cjd_flows.transforms import ( +from cjd_flows.transforms import ( ScaleTransform, MaskedCoupling, LUTransform, diff --git a/src/cjd_flows/networks.py b/src/cjd_flows/networks.py index c9494ff..def1827 100644 --- a/src/cjd_flows/networks.py +++ b/src/cjd_flows/networks.py @@ -998,7 +998,7 @@ class JetConditioner(nn.Module): Kolesnikov et al. 2024). Shape-preserving network intended as the conditioner of a - :class:`~src.cjd_flows.transforms.MaskedCoupling` layer: since additive + :class:`~cjd_flows.transforms.MaskedCoupling` layer: since additive coupling has unit Jacobian determinant irrespective of the conditioner, arbitrary capacity (including global self-attention) can be spent here without affecting the constant-Jacobian-determinant property of the flow. diff --git a/tests/cjd_flows/flows_test.py b/tests/cjd_flows/flows_test.py index c72ab55..989fcc3 100644 --- a/tests/cjd_flows/flows_test.py +++ b/tests/cjd_flows/flows_test.py @@ -1,8 +1,8 @@ import torch from pyro.distributions import Normal -from src.cjd_flows.flows import USFlow -from src.cjd_flows.distributions import RadialDistribution, GammaMM +from cjd_flows.flows import USFlow +from cjd_flows.distributions import RadialDistribution, GammaMM from pyro.nn import DenseNN def test_onnx(): diff --git a/tests/cjd_flows/linalg_test.py b/tests/cjd_flows/linalg_test.py index 0fde5eb..24865a3 100644 --- a/tests/cjd_flows/linalg_test.py +++ b/tests/cjd_flows/linalg_test.py @@ -1,5 +1,5 @@ import torch -from src.cjd_flows.linalg import solve_triangular +from cjd_flows.linalg import solve_triangular test_size = 10 tol = 1e-5 diff --git a/tests/cjd_flows/networks_test.py b/tests/cjd_flows/networks_test.py index 6c0c5c5..cc30f91 100644 --- a/tests/cjd_flows/networks_test.py +++ b/tests/cjd_flows/networks_test.py @@ -1,7 +1,7 @@ import torch -from src.cjd_flows.networks import JetConditioner, sincos_pos_embed -from src.cjd_flows.transforms import MaskedCoupling +from cjd_flows.networks import JetConditioner, sincos_pos_embed +from cjd_flows.transforms import MaskedCoupling def test_sincos_pos_embed(): diff --git a/tests/cjd_flows/transforms_test.py b/tests/cjd_flows/transforms_test.py index 6453303..2f9882a 100644 --- a/tests/cjd_flows/transforms_test.py +++ b/tests/cjd_flows/transforms_test.py @@ -1,7 +1,7 @@ import torch import math -from src.cjd_flows.transforms import ( +from cjd_flows.transforms import ( ScaleTransform, Permute, LUTransform, diff --git a/tests/explib/hyperopt_test.py b/tests/explib/hyperopt_test.py index 8a3740d..9552db9 100644 --- a/tests/explib/hyperopt_test.py +++ b/tests/explib/hyperopt_test.py @@ -1,7 +1,7 @@ import os import typing as T -from src.cjd_flows.explib.config_parser import read_config +from cjd_flows.explib.config_parser import read_config def test_mnist(): diff --git a/tests/explib/mnist.yaml b/tests/explib/mnist.yaml index f3bcd1b..fb1018c 100644 --- a/tests/explib/mnist.yaml +++ b/tests/explib/mnist.yaml @@ -1,9 +1,9 @@ --- -__object__: src.cjd_flows.explib.base.ExperimentCollection +__object__: cjd_flows.explib.base.ExperimentCollection name: mnist_ablation_best_veriflow experiments: - &exp_laplace0 - __object__: src.cjd_flows.explib.hyperopt.HyperoptExperiment + __object__: cjd_flows.explib.hyperopt.HyperoptExperiment name: mnist0 scheduler: __object__: ray.tune.schedulers.ASHAScheduler @@ -25,7 +25,7 @@ experiments: image_shape: [28, 28] dataset: class: - __class__: src.cjd_flows.explib.datasets.MnistSplit + __class__: cjd_flows.explib.datasets.MnistSplit params: dataloc: /home/faried/Projects/USFlows/data/mnist space_to_depth_factor: 4 @@ -35,14 +35,14 @@ experiments: __eval__: tune.choice([32]) optim_cfg: optimizer: - __class__: src.cjd_flows.sophia.SophiaG + __class__: cjd_flows.sophia.SophiaG params: lr: __eval__: 1e-3 weight_decay: 0.0 model_cfg: type: - __class__: src.cjd_flows.flows.USFlow + __class__: cjd_flows.flows.USFlow params: soft_training: __eval__: tune.choice([False]) @@ -57,7 +57,7 @@ experiments: lu_transform: 1 householder: 1 conditioner_cls: - __class__: src.cjd_flows.networks.ConvNet2D + __class__: cjd_flows.networks.ConvNet2D conditioner_args: c_in: 16 c_hidden: @@ -76,14 +76,14 @@ experiments: nonlinearity: __eval__: tune.choice([torch.nn.ReLU()]) base_distribution: - __object__: src.cjd_flows.distributions.RadialDistribution + __object__: cjd_flows.distributions.RadialDistribution device: cpu p: __eval__: float("1") loc: __eval__: torch.zeros([16, 7, 7]).to("cpu") norm_distribution: - __object__: src.cjd_flows.distributions.GammaMM + __object__: cjd_flows.distributions.GammaMM concentration: __eval__: torch.rand([50]).to("cpu") * 75 rate: