added overfit test, updated documentation, fixed yapf and flake8 erro…

…rs (deepchem#4184)

* added overfit test, updated documentation, fixed yapf and flake8 errors

* minor fix

deepchem/data/tests/test_deepvariant_pileup_featurizer.py

@@ -27,7 +27,7 @@ def test_pileup(self):
 
         # Assert the number of reads
         self.assertEqual(len(image_dataset), 15)
-        self.assertEqual(image_dataset.X[0].shape, (100, 221, 6))
+        self.assertEqual(image_dataset.X[0].shape, (299, 299, 6))
 
 
 if __name__ == "__main__":

deepchem/feat/deepvariant_featurizer.py

@@ -6,6 +6,7 @@
 try:
     import dgl
     import torch
+    import pysam
 except ImportError:
     pass
 
@@ -606,8 +607,8 @@ def seq_to_int(seq):
 
     def fast_pass_aligner(self, assembled_region: Dict[str, Any]) -> List[Any]:
         """
-        Align reads to the haplotype of the assembled region using Striped Smith
-        Waterman algorithm.
+        Align reads to the haplotype of the assembled region using Striped
+        Smith Waterman algorithm.
 
         Parameters
         ----------
@@ -807,7 +808,8 @@ def _featurize(self, datapoint):
             decoded_sequences.append(decoded_seq)
 
         # Map the sequences to chrom names
-        chrom_names = ["chr1", "chr2"]
+        with pysam.FastaFile(reference_file_path) as fasta_file:
+            chrom_names = fasta_file.references
 
         reference_seq_dict = {
             chrom_names[i]: seq for i, seq in enumerate(decoded_sequences)

deepchem/feat/deepvariant_pileup_featurizer.py

@@ -3,6 +3,11 @@
 from deepchem.data import ImageDataset
 from typing import List
 
+try:
+    import pysam
+except ImportError:
+    pass
+
 
 class PileupFeaturizer(Featurizer):
     """
@@ -23,13 +28,16 @@ class PileupFeaturizer(Featurizer):
     >>> from deepchem.feat import RealignerFeaturizer, PileupFeaturizer
     >>> bamfile_path = 'deepchem/data/tests/example.bam'
     >>> reference_path = 'deepchem/data/tests/sample.fa'
-    >>> realigner_feat = RealignerFeaturizer()
-    >>> windows_haplotypes = realigner_feat.featurize((bamfile_path, reference_path))
+    >>> realigner= RealignerFeaturizer()
+    >>> windows_haplotypes = realigner.featurize((bamfile_path,reference_path))
     >>> pileup_feat = PileupFeaturizer()
     >>> features = pileup_feat.featurize((windows_haplotypes, reference_path))
-    >>> features
-    <ImageDataset X.shape: (15, 100, 221, 6), y.shape: (15,), w.shape: (15,), ids: [0 1 2 ... 12 13 14], task_names: [0]>
 
+    Note
+    ----
+    This class requires pysam to be installed. Pysam can be used with
+    Linux or MacOS X. To use Pysam on Windows, use Windows Subsystem for
+    Linux(WSL).
 
     """
 
@@ -91,7 +99,8 @@ def _featurize(self, datapoint):
             decoded_sequences.append(decoded_seq)
 
         # Map the sequences to chrom names
-        chrom_names = ["chr1", "chr2"]
+        with pysam.FastaFile(reference_file_path) as fasta_file:
+            chrom_names = fasta_file.references
 
         reference_seq_dict = {
             chrom_names[i]: seq for i, seq in enumerate(decoded_sequences)
@@ -132,8 +141,8 @@ def get_supports_variant_intensity(read, haplotype):
         def get_diff_from_ref_intensity(base, ref_base):
             return 1.0 if base != ref_base else 0.25
 
-        height = 221
-        width = 100
+        height = 299
+        width = 299
         num_channels = 6
 
         images = []

deepchem/models/torch_models/inceptionv3.py

@@ -11,13 +11,16 @@
 
 
 class InceptionV3(nn.Module):
+    """
+    InceptionV3 model architecture for image classification.
+    """
 
     def __init__(self,
-                 num_classes=1000,
-                 aux_logits=True,
+                 num_classes: int = 1000,
+                 aux_logits: bool = True,
                  transform_input: bool = False,
-                 in_channels=6,
-                 dropout_rate=0.5):
+                 in_channels: int = 6,
+                 dropout_rate: float = 0.5) -> None:
         super(InceptionV3, self).__init__()
         self.aux_logits = aux_logits
 
@@ -85,7 +88,7 @@ def __init__(self,
                 nn.init.constant_(module.weight, 1)
                 nn.init.constant_(module.bias, 0)
 
-    def forward(self, x):
+    def forward(self, x: torch.Tensor):
         # N x 6 x 299 x 299
         x = self.Conv2d_1a_3x3(x)
         # N x 32 x 149 x 149
@@ -770,16 +773,16 @@ class InceptionV3Model(TorchModel):
     def __init__(self,
                  in_channels=6,
                  warmup_steps=10000,
-                 learning_rate=0.001,
+                 learning_rate=0.064,
+                 dropout_rate=0.2,
+                 decay_rate=0.94,
                  **kwargs):
         # Fixed hyperparameters
         decay_steps = 2  # epochs per decay
-        decay_rate = 0.947
         rho = 0.9
         momentum = 0.9
         epsilon = 1.0
         # weight_decay = 0.00004
-        dropout_rate = 0.2
 
         # Initialize the InceptionV3 model architecture
         model = InceptionV3(num_classes=3,

deepchem/models/torch_models/tests/test_inceptionv3.py

@@ -241,3 +241,51 @@ def test_InceptionAux():
 
     # Verify the output shape matches expected shape
     assert output_tensor.shape == expected_output_shape
+
+
+@pytest.mark.torch
+def test_inceptionv3_overfit():
+    """
+    Test the InceptionV3 model's ability to overfit on a small dataset
+    """
+    from sklearn.metrics import accuracy_score
+    from deepchem.models.torch_models.inceptionv3 import InceptionV3Model
+
+    # Generate a small dataset to test overfitting
+    input_shape = (3, 3, 299, 299)
+    input_samples = np.random.randn(*input_shape).astype(np.float32)
+    output_samples = np.array([0, 1,
+                               2]).astype(np.int64)  # One sample per class
+    one_hot_output_samples = one_hot_encode(output_samples, 3)
+
+    dataset = dc.data.ImageDataset(input_samples, one_hot_output_samples)
+
+    # Initialize model and set a temporary directory for saving
+    model_dir = tempfile.mkdtemp()
+    inception_model = InceptionV3Model(n_tasks=3,
+                                       in_channels=3,
+                                       warmup_steps=0,
+                                       learning_rate=0.1,
+                                       decay_rate=1,
+                                       dropout_rate=0.0,
+                                       model_dir=model_dir)
+
+    # Train for many epochs to test overfitting capability
+    inception_model.fit(dataset, nb_epoch=100)
+
+    # Check performance on the small dataset
+    pred = inception_model.predict(dataset)
+
+    # Ensure predictions are in the correct shape to match the number
+    # of classes
+    assert pred.shape == (3, 3), f"Unexpected prediction shape: {pred.shape}"
+
+    # Convert predictions and labels to one-hot format for metric computation
+    pred_labels = np.argmax(pred, axis=1)
+    true_labels = output_samples
+
+    # Calculate accuracy using direct comparison
+    accuracy = accuracy_score(true_labels, pred_labels)
+
+    # Assert the accuracy is high, indicating overfitting
+    assert accuracy > 0.9, "Failed to overfit on small dataset"