Implemented outlier interpolation

deepof/data.py

@@ -836,7 +836,7 @@ class coordinates:
 
         """
 
-        trained_models = deepof.train_utils.deep_unsupervised_embedding(
+        trained_models = deepof.train_utils.autoencoder_fitting(
             preprocessed_object=preprocessed_object,
             batch_size=batch_size,
             encoding_size=encoding_size,

deepof/train_utils.py

@@ -139,7 +139,7 @@ def get_callbacks(
     return callbacks
 
 
-def deep_unsupervised_embedding(
+def autoencoder_fitting(
     preprocessed_object: Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray],
     batch_size: int,
     encoding_size: int,
@@ -308,6 +308,9 @@ def deep_unsupervised_embedding(
                 ],
             )
 
+            if save_weights:
+                ae.save_weights("{}_final_weights.h5".format(run_ID))
+
         else:
 
             callbacks_ = cbacks + [
@@ -340,7 +343,7 @@ def deep_unsupervised_embedding(
             ae.fit(
                 x=Xs,
                 y=ys,
-                epochs=35,
+                epochs=2,
                 batch_size=batch_size,
                 verbose=1,
                 validation_data=(
@@ -350,6 +353,9 @@ def deep_unsupervised_embedding(
                 callbacks=callbacks_,
             )
 
+            if save_weights:
+                ae.save_weights("{}_final_weights.h5".format(run_ID))
+
             if log_hparams:
                 # noinspection PyUnboundLocalVariable
                 def tensorboard_metric_logging(run_dir: str, hpms: Any):

tests/test_train_utils.py

@@ -142,7 +142,7 @@ def test_tune_search(
         loss=loss,
         mmd_warmup_epochs=0,
         overlap_loss=overlap_loss,
-        pheno_class=pheno_class,
+        phenotype_class=pheno_class,
         predictor=predictor,
         project_name="test_run",
         callbacks=callbacks,