Merge pull request #10 from discovery-unicamp/adding_options

Adding validation options

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "librep"
-version = "0.0.2-dev"
+version = "0.0.3-dev"
 description = "HIAAC-M4 representation learning framework"
 readme = "README.md"
 authors = [{name = "Otavio Napoli", email = "otavio.napoli@gmail.com"}]
@@ -57,5 +57,5 @@ deep-learning = [
 ]
 
 [project.urls]
-"Homepage" = "https://github.com/otavioon/hiaac-librep"
-"Bug Tracker" = "https://github.com/otavioon/hiaac-librep/issues"
+"Homepage" = "https://github.com/discovery-unicamp/hiaac-librep"
+"Bug Tracker" = "https://github.com/discovery-unicamp/hiaac-librep/issues"

src/librep/base/__init__.py

@@ -0,0 +1,20 @@
+from librep.base.estimator import Estimator
+from librep.base.transform import Transform, InvertibleTransform
+from librep.base.evaluators import (
+    SupervisedEvaluator,
+    UnsupervisedEvaluator,
+    CustomSimpleEvaluator,
+    CustomMultiEvaluator,
+    Evaluators,
+)
+
+__all__ = [
+    "Estimator",
+    "Transform",
+    "InvertibleTransform",
+    "SupervisedEvaluator",
+    "UnsupervisedEvaluator",
+    "CustomSimpleEvaluator",
+    "CustomMultiEvaluator",
+    "Evaluators",
+]

src/librep/base/estimator.py

@@ -5,23 +5,38 @@
 # Wrap around scikit learn base API
 # Borrowed from Sklearn API
 
-class Estimator(Parametrizable):
-    """An object which manages the estimation and decoding of a model.
-    """
 
-    def fit(self, X: ArrayLike, y: ArrayLike = None, **estimator_params) -> 'Estimator':
+class Estimator(Parametrizable):
+    """An object which manages the estimation and decoding of a model."""
+
+    def fit(
+        self,
+        X: ArrayLike,
+        y: ArrayLike = None,
+        X_val: ArrayLike = None,
+        y_val: ArrayLike = None,
+        **estimator_params,
+    ) -> "Estimator":
         """Fits the model function arround X. It takes some samples (X) and
         the respective labels (y) if the model is supervised.
+        X_val and y_val are optional and may be used if needed.
 
         Parameters
         ----------
         X : ArrayLike
-            An array-like representing the whole dataset with shape:
+            An array-like representing the whole training dataset with shape:
             (n_samples, n_features).
         y : ArrayLike
-            The respective labels, with shape: (n_samples, ). This parameter is
-            optional and may be used if needed.
-        **estimator_params : type
+            The respective train labels, with shape: (n_samples, ). This parameter is
+            optional and may be used if needed. By default None.
+        X_val : ArrayLike
+            An array-like representing the whole validation dataset with shape:
+            (k_samples, n_features). This parameter is optional and may be used
+            if needed. By default None.
+        y_val : ArrayLike
+            The respective validation labels, with shape: (k_samples, ). This
+            parameter is optional and may be used if needed. By default None.
+        **estimator_params : Any
             Optional data-dependent parameters.
 
         Returns
@@ -30,8 +45,7 @@ def fit(self, X: ArrayLike, y: ArrayLike = None, **estimator_params) -> 'Estimat
             The estimator (self).
 
         """
-
-        raise NotImplementedError
+        return self
 
     def predict(self, X: ArrayLike) -> ArrayLike:
         """Predict or project the values of the samples (X).
@@ -51,21 +65,32 @@ def predict(self, X: ArrayLike) -> ArrayLike:
 
         raise NotImplementedError
 
-    def fit_predict(self,
-                    X: ArrayLike,
-                    y: ArrayLike = None,
-                    **estimator_params) -> ArrayLike:
+    def fit_predict(
+        self,
+        X: ArrayLike,
+        y: ArrayLike = None,
+        X_val: ArrayLike = None,
+        y_val: ArrayLike = None,
+        **estimator_params,
+    ) -> ArrayLike:
         """Chain fit and predict methods, togheter. It fits the model and
         predict the samples of X.
 
         Parameters
         ----------
         X : ArrayLike
-            An array-like representing the whole dataset with shape:
+            An array-like representing the whole training dataset with shape:
             (n_samples, n_features).
         y : ArrayLike
-            The respective labels, with shape: (n_samples, ). This parameter is
-            optional and may be used if needed.
+            The respective training labels, with shape: (n_samples, ). This parameter is
+            optional and may be used if needed. By default None.
+        X_val : ArrayLike
+            An array-like representing the whole validation dataset with shape:
+            (k_samples, n_features). This parameter is optional and may be used
+            if needed. By default None.
+        y_val : ArrayLike
+            The respective validation labels, with shape: (k_samples, ). This
+            parameter is optional and may be used if needed. By default None.
         **estimator_params : type
             Optional data-dependent parameters.
 
@@ -76,5 +101,5 @@ def fit_predict(self,
             labels predicted for each samples.
 
         """
-        self.fit(X, y, **estimator_params)
+        self.fit(X, y, X_val, y_val, **estimator_params)
         return self.predict(X)

src/librep/base/transform.py

@@ -5,13 +5,21 @@
 # Wrap around scikit learn base API
 # Borrowed from Sklearn API
 
+
 class Transform(Parametrizable):
     """For filtering or modifying the data, in a supervised or unsupervised way.
     `fit` allows implementing parametrizable transforms. This method sees the
     whole dataset. `transform` allows transforming each sample.
     """
 
-    def fit(self, X: ArrayLike, y: ArrayLike = None, **fit_params) -> 'Transform':
+    def fit(
+        self,
+        X: ArrayLike,
+        y: ArrayLike = None,
+        X_val: ArrayLike = None,
+        y_val: ArrayLike = None,
+        **fit_params,
+    ) -> "Transform":
         """Fit the transformation with information of the whole dataset.
 
         Parameters
@@ -22,6 +30,13 @@ def fit(self, X: ArrayLike, y: ArrayLike = None, **fit_params) -> 'Transform':
         y : ArrayLike
             The respective labels, with shape: (n_samples, ). This parameter is
             optional and may be used if needed.
+        X_val : ArrayLike
+            An array-like representing the whole validation dataset with shape:
+            (k_samples, n_features). This parameter is optional and may be used
+            if needed.
+        y_val : ArrayLike
+            The respective validation labels, with shape: (k_samples, ). This
+            parameter is optional and may be used if needed.
         **fit_params : type
             Optional data-dependent parameters.
 
@@ -49,8 +64,39 @@ def transform(self, X: ArrayLike) -> ArrayLike:
         """
         raise NotImplementedError
 
-    def fit_transform(self, X: ArrayLike, y: ArrayLike = None, **fit_params):
-        self.fit(X, y, **fit_params)
+    def fit_transform(
+        self,
+        X: ArrayLike,
+        y: ArrayLike = None,
+        X_val: ArrayLike = None,
+        y_val: ArrayLike = None,
+        **fit_params,
+    ) -> ArrayLike:
+        """Chain fit and transform methods, toghether. It firs the model and
+        then transforms the training data.
+
+        Parameters
+        ----------
+        X : ArrayLike
+            An array-like representing the whole dataset with shape:
+            (n_samples, n_features).
+        y : ArrayLike
+            The respective train labels, with shape: (n_samples, ). This parameter is
+            optional and may be used if needed. By default None.
+        X_val : ArrayLike
+            An array-like representing the whole validation dataset with shape:
+            (k_samples, n_features). This parameter is optional and may be used
+            if needed. By default None.
+        y_val : ArrayLike
+            The respective validation labels, with shape: (k_samples, ). This
+            parameter is optional and may be used if needed. By default None.
+
+        Returns
+        -------
+        ArrayLike
+            An array-like with the transformed samples.
+        """
+        self.fit(X, y, X_val, y_val, **fit_params)
         return self.transform(X)
 
 

src/librep/config/type_definitions.py

@@ -1,8 +1,9 @@
 import os
 import numpy
-from typing import Union, Hashable
+import pandas as pd
+from typing import Union, Hashable, Iterable
 
 # PathLike: The PathLike type is used for defining a file path.
 PathLike = Union[str, os.PathLike]
-ArrayLike = Union[numpy.ndarray]
+ArrayLike = Union[numpy.ndarray, pd.DataFrame, Iterable]
 KeyType = Hashable