diff --git a/diffxpy/pkg_constants.py b/diffxpy/pkg_constants.py
index 12dfea2..48cabe0 100644
--- a/diffxpy/pkg_constants.py
+++ b/diffxpy/pkg_constants.py
@@ -15,4 +15,4 @@
BATCHGLM_BACKEND = "tf1"
BATCHGLM_FEATUREWISE = True
-BATCHGLM_AUTOGRAD = True
+BATCHGLM_AUTOGRAD = False
diff --git a/diffxpy/testing/tests.py b/diffxpy/testing/tests.py
index c378494..a8c9445 100644
--- a/diffxpy/testing/tests.py
+++ b/diffxpy/testing/tests.py
@@ -194,6 +194,8 @@ def _fit(
constructor_args = {}
if quick_scale is not None:
constructor_args["quick_scale"] = quick_scale
+ if batch_size is not None and backend != "tf2":
+ constructor_args["batch_size"] = batch_size
# Backend-specific constructor arguments:
if backend.lower() in ["tf1"]:
constructor_args['provide_optimizers'] = {
diff --git a/diffxpy/unit_test/test_compare_numpy_tf2.py b/diffxpy/unit_test/test_compare_numpy_tf2.py
new file mode 100644
index 0000000..8ddb4f1
--- /dev/null
+++ b/diffxpy/unit_test/test_compare_numpy_tf2.py
@@ -0,0 +1,81 @@
+import unittest
+import logging
+import diffxpy.api as de
+import numpy as np
+
+
+class TestBackendsNb(unittest.TestCase):
+
+ """
+ Negative binomial noise model unit tests that test whether the wald test results
+ in the same logfoldchange, p- and q-values and the coefficents are the same after
+ fitting when using the same simulated data.
+ """
+
+ def __init__(self, *args, **kwargs):
+
+ super(TestBackendsNb, self).__init__(*args, **kwargs)
+
+ from batchglm.api.models.numpy.glm_nb import Simulator
+ self.sim = Simulator(num_observations=10000, num_features=200)
+ self.sim.generate_sample_description(num_batches=0, num_conditions=4)
+ self.sim.generate_params()
+ self.sim.generate_data()
+
+ logging.getLogger("tensorflow").setLevel(logging.ERROR)
+ logging.getLogger("batchglm").setLevel(logging.WARNING)
+ logging.getLogger("diffxpy").setLevel(logging.WARNING)
+
+ self.numpy_results = de.test.wald(
+ data=self.sim.input_data,
+ sample_description=self.sim.sample_description,
+ factor_loc_totest="condition",
+ formula_loc="~ 1 + condition + batch",
+ noise_model='nb',
+ backend='numpy'
+ )
+ _ = self.numpy_results.summary()
+
+ self.tf2_results = de.test.wald(
+ data=self.sim.input_data,
+ sample_description=self.sim.sample_description,
+ factor_loc_totest="condition",
+ formula_loc="~ 1 + condition + batch",
+ noise_model='nb',
+ backend='tf2'
+ )
+ _ = self.tf2_results.summary()
+
+ """
+ Test with numpy:
+ """
+
+ def test_coeff_similarity(self):
+
+ a_var_max_diff = np.max(np.abs(self.tf2_results.model_estim.a_var-self.numpy_results.model_estim.a_var))
+ b_var_max_diff = np.max(np.abs(self.tf2_results.model_estim.b_var-self.numpy_results.model_estim.b_var))
+ assert a_var_max_diff < 1e-6 and b_var_max_diff < 1e-6, \
+ ("a_var_max_diff: %f, b_var_max_diff: %f", a_var_max_diff, b_var_max_diff)
+
+ return True
+
+ def test_logfoldchange_similarity(self):
+ max_diff = np.max(np.abs(self.tf2_results.summary()['log2fc'].values-self.numpy_results.summary()['log2fc'].values))
+ assert max_diff < 1e-12, ("log_fold_change difference: %f > 1e-12", max_diff)
+
+ return True
+
+ def test_pval_similarity(self):
+ max_diff = np.max(np.abs(self.tf2_results.pval-self.numpy_results.pval))
+ assert max_diff < 1e-12, ("p-val difference: %f > 1e-12", max_diff)
+
+ return True
+
+ def test_qval_similarity(self):
+ max_diff = np.max(np.abs(self.tf2_results.summary()['qval'].values-self.numpy_results.summary()['qval'].values))
+ assert max_diff < 1e-12, ("q-val difference: %f > 1e-12", max_diff)
+
+ return True
+
+if __name__ == '__main__':
+ unittest.main()
diff --git a/diffxpy/unit_test/test_fit.py b/diffxpy/unit_test/test_fit.py
index 99c5430..39d3bce 100644
--- a/diffxpy/unit_test/test_fit.py
+++ b/diffxpy/unit_test/test_fit.py
@@ -17,7 +17,7 @@ def _test_model_fit(
"""
Test if de.wald() generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
- of the two-side Kolmgorov-Smirnov test for equality of the observed
+ of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distribution and a uniform distribution.
:param n_cells: Number of cells to simulate (number of observations per test).
@@ -34,7 +34,7 @@ def _test_model_fit(
raise ValueError("noise model %s not recognized" % noise_model)
sim = Simulator(num_observations=n_cells, num_features=n_genes)
- sim.generate_sample_description(num_batches=0, num_conditions=0)
+ sim.generate_sample_description(num_batches=1, num_conditions=1)
sim.generate_params(rand_fn_scale=rand_fn_scale)
sim.generate_data()
@@ -77,7 +77,7 @@ def _test_model_fit_partition(
raise ValueError("noise model %s not recognized" % noise_model)
sim = Simulator(num_observations=n_cells, num_features=n_genes)
- sim.generate_sample_description(num_batches=0, num_conditions=0)
+ sim.generate_sample_description(num_batches=1, num_conditions=1)
sim.generate_params(rand_fn_scale=rand_fn_scale)
sim.generate_data()
@@ -121,7 +121,7 @@ def _test_residuals_fit(
raise ValueError("noise model %s not recognized" % noise_model)
sim = Simulator(num_observations=n_cells, num_features=n_genes)
- sim.generate_sample_description(num_batches=0, num_conditions=0)
+ sim.generate_sample_description(num_batches=1, num_conditions=1)
sim.generate()
random_sample_description = pd.DataFrame({
@@ -209,23 +209,23 @@ def test_residuals_fit(
noise_model="nb"
)
-
+"""
class TestFitNorm(_TestFit, unittest.TestCase):
- """
+
Normal noise model unit tests that tests whether model fit relay works.
- """
+
def test_model_fit(
self,
n_cells: int = 2000,
n_genes: int = 2
):
- """
+
Test if model fit for "norm" noise model works.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
- """
+
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
@@ -242,12 +242,12 @@ def test_model_fit_partition(
n_cells: int = 2000,
n_genes: int = 2
):
- """
+
Test if partitioned model fit for "norm" noise model works.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
- """
+
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
@@ -264,12 +264,12 @@ def test_residuals_fit(
n_cells: int = 2000,
n_genes: int = 2
):
- """
+
Test if residual fit for "norm" noise model works.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
- """
+
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
@@ -280,7 +280,7 @@ def test_residuals_fit(
n_genes=n_genes,
noise_model="norm"
)
-
+"""
if __name__ == '__main__':
unittest.main()