Merge pull request #106 from maxsch3/feature/105

Feature/105

Author: maxsch3

build_requirements.txt

@@ -1,6 +1,7 @@
 setuptools-scm==3.5.0
-mkdocs==1.0.4
-mkdocs-jupyter==0.10.2
+mkdocs==1.1.2
+mkdocs-jupyter==0.13.0
 notebook==6.0.2
 pymdown-extensions==6.3
+nbconvert==5.6.1
 git+https://github.com/tomchristie/mkautodoc.git#egg=mkautodoc

keras_batchflow/base/batch_shapers/var_shaper.py

@@ -5,6 +5,19 @@
 
 class VarShaper:
 
+    """
+    This class is a wrapper around encoder. It abstracts away the encoding of a single column into a
+    model-ready data. This class is used in BatchShaper where all leaves (a tuple (column name, encoder))
+    in a structure are replaced with equivalent VarShaper objects in a BatchShaper's constructor method
+    After that, BatchShaper use this equivalent structure of VarShaper objects rather than original
+    "human-readable" structure of tuples.
+
+    This class is a backend class and you normally would not need to use it manually.
+
+    It implements core interface functions like transform, inverse_transform,
+    as well as additional metadata functions like, shape, n_classes, dtype, etc.
+    """
+
     _dummy_constant_counter = 0
 
     def __init__(self, var_name, encoder, data_sample=None):
@@ -109,7 +122,6 @@ def _get_dtypes(self, sample):
         else:
             RuntimeError(f"Error: the class type {self._encoder} is not supported in '_get_dtypes' method")
 
-
     def _get_n_classes(self, encoder):
         """
         Calculates number of classes provided by the encoder. This is required for creating embedding layer for

keras_batchflow/base/batch_transformers/base_random_cell.py

@@ -326,3 +326,6 @@ def transform(self, batch):
         else:
             batch[self._cols] = transformed
         return batch
+
+    def inverse_transform(self, batch):
+        return batch

keras_batchflow/base/batch_transformers/batch_fork.py

@@ -17,3 +17,6 @@ def _verify_levels(self, levels):
     def transform(self, batch):
         batch = pd.concat([batch]*len(self._levels), axis=1, keys=self._levels)
         return batch
+
+    def inverse_transform(self, batch):
+        return batch

keras_batchflow/base/batch_transformers/batch_transformer.py

@@ -1,11 +1,17 @@
+from abc import ABC, abstractmethod
 
 
-class BatchTransformer:
+class BatchTransformer(ABC):
     """
     This is an abstract class that defines basic functionality and interfaces of all BatchTransformers
     """
     def __init__(self):
         pass
 
+    @abstractmethod
     def transform(self, batch):
-        return batch
+        pass
+
+    @abstractmethod
+    def inverse_transform(self, batch):
+        pass

test/test_batch_generator.py

@@ -132,7 +132,18 @@ def transform(self, batch):
                 batch[self.col_name] = 'Red'
                 return batch
 
-        bt1 = BatchTransformer()
+            def inverse_transform(self, batch):
+                return batch
+
+        class TransparentTransform(BatchTransformer):
+
+            def transform(self, batch):
+                return batch
+
+            def inverse_transform(self, batch):
+                return batch
+
+        bt1 = TransparentTransform()
         bt2 = TestTransform('label')
         bg = BatchGenerator(
             self.df,

test/test_triplet_pk_generator.py

@@ -54,9 +54,17 @@ def test_basic(self):
             assert batch[0].shape == (6, 1)
             assert batch[1].shape == (6, 1)
 
-
     def test_kwargs_pass_to_parent(self):
-        bt = BatchTransformer()
+
+        class TransparentTransform(BatchTransformer):
+
+            def transform(self, batch):
+                return batch
+
+            def inverse_transform(self, batch):
+                return batch
+
+        bt = TransparentTransform()
         tg = TripletPKGenerator(
             data=self.df,
             triplet_label='label',