nice fmri fig

Author: csinva

intermediate/readme.md renamed to data/intermediate/readme.md

@@ -0,0 +1,11 @@
+figure.autolayout : True
+
+axes.titlesize : 20
+axes.spines.top:    False
+axes.spines.right:  False
+axes.labelsize      : 15  ## fontsize of the x any y labels
+
+font.size           : 15
+
+xtick.labelsize      : 15 ## fontsize of the tick labels
+ytick.labelsize      : 15 ## fontsize of the tick labels

fmri/analyze_fmri.ipynb

@@ -68,13 +68,13 @@ def get_roi_and_idx(run):
         runs = [int(sys.argv[-1])]
     else:
         runs = list(range(300)) # this number determines which neuron we will pick
-    print('runs', runs)
+    print('\nruns', runs)
 
     # fit linear models
     use_sigmas = False
     use_small = False
     out_dir = '/scratch/users/vision/data/gallant/vim_2_crcns'
-    save_dir = oj(out_dir, 'dec13_baselines')
+    save_dir = oj(out_dir, 'dec14_baselines_ard')
     suffix = '_feats' # _feats, '' for pixels
     norm = '_norm' # ''
     print('saving to', save_dir)
@@ -140,14 +140,17 @@ def get_roi_and_idx(run):
 
         # only fit voxels with no missing vals
         if not (n_train == y_train.size and num_test == y_test.size):
+            print('\tskipping this voxel!')
             continue
 
         # reg values to try
         reg_params = np.logspace(3, 6, 20).round().astype(int)
 
         # fit ard + mdl-rs
         baselines = {}
-        for model_type, model_name in zip([ARDRegression, RidgeULNML], ['ard', 'mdl-rs']):
+        for model_type, model_name in zip([ARDRegression], ['ard']):
+#         for model_type, model_name in zip([ARDRegression, RidgeULNML], ['ard', 'mdl-rs']):
+            print('\tfitting', model_name)
             model = model_type()
             model.fit(X_train, y_train)
             preds_train = model.predict(X_train)
@@ -159,6 +162,7 @@ def get_roi_and_idx(run):
             baselines[f'{model_name}_corr'] = np.corrcoef(y_test, preds)[0, 1]
 
         # fit ridge cv
+        print('\tfitting ridgecv...')
         m = RidgeCV(alphas=reg_params, store_cv_values=True)
         m.fit(X_train, y_train)
         preds_train = m.predict(X_train)
@@ -168,7 +172,7 @@ def get_roi_and_idx(run):
         mse = metrics.mean_squared_error(y_test, preds)
         r2 = metrics.r2_score(y_test, preds)
         corr = np.corrcoef(y_test, preds)[0, 1]
-        print('RidgeCV corr', corr)
+        print('\tRidgeCV corr', corr)
 
 
         # fit mdl comp
@@ -240,4 +244,5 @@ def get_roi_and_idx(run):
             **r,
             **baselines,
         }
-        pkl.dump(results, open(oj(save_dir, f'ridge_{i}.pkl'), 'wb'))
+        pkl.dump(results, open(oj(save_dir, f'ridge_{i}.pkl'), 'wb'))
+        print('\tdone!')

fmri/matplotlibrc

@@ -3,6 +3,7 @@
 from numpy import ndarray
 import scipy.linalg as linalg
 from typing import Union, Tuple
+from tqdm import tqdm
 import warnings
 
 from sklearn.base import BaseEstimator
@@ -51,7 +52,7 @@ def fit(self, X: ndarray, y: ndarray) -> 'RidgeULNML':
 
         self.call_before_fit(X, y)
 
-        for i in range(self.n_iter):
+        for i in tqdm(range(self.n_iter)):
             self.beta_ = self.fit_beta(C, b, self.lam_)
             self.sigma2_ = self.fit_sigma2(X, y, self.beta_, self.lam_)
             self.lam_ = self.fit_lam(

fmri/run.py

@@ -4,9 +4,9 @@ Official code for using / reproducing MDL-COMP from the paper "Revisiting comple
 
 # Reproducing the results in the paper
 - most of the results can be produced by simply running the notebooks
-- the experiments with real-data are more in depth and require running the `submit_real_data_jobs.py` file (which is a script that calls `fit.py` with the appropriate hyperparameters) before running the notebook to view the analysis
+- the experiments with real-data are more in depth and require running `scripts/submit_real_data_jobs.py` (which is a script that calls `src/fit.py` with the appropriate hyperparameters) before running the notebook to view the analysis
 
-![](https://csinva.github.io/mdl-complexity/results/fig_iid_mse.svg)
+![](https://csinva.github.io/mdl-complexity/reports/fig_iid_mse.svg)
 
 
 ## Calculating MDL-COMP

intermediate/mdl_results_d_100_200.txt

@@ -10,7 +10,7 @@
 
 
 # run
-s = Slurm("fmri", {"partition": partition, "time": "1-0"})
+s = Slurm("fmri", {"partition": partition, "time": "2-0"})
 ks = sorted(params_to_vary.keys())
 vals = [params_to_vary[k] for k in ks]
 param_combinations = list(itertools.product(*vals)) # list of tuples
@@ -19,7 +19,7 @@
 
 # iterate
 for i in range(len(param_combinations)):
-    param_str = 'module load python; python3 run.py '
+    param_str = 'module load python; python3 ../fmri/run.py '
     for j, key in enumerate(ks):
         param_str += key + ' ' + str(param_combinations[i][j]) + ' '
     print(param_str)

intermediate/ridge_identity_bias_variance.txt

@@ -137,7 +137,7 @@
 
     # iterate
     for i in range(len(params_full)):
-        param_str = 'python3 fit.py '
+        param_str = 'python3 ../src/fit.py '
         for j, key in enumerate(ks):
             param_str += key + ' ' + str(params_full[i][j]) + ' '
         subprocess.call(param_str, shell=True)