zeroth-learn 0.2.4__tar.gz → 0.2.5__tar.gz

{zeroth_learn-0.2.4/zeroth_learn.egg-info → zeroth_learn-0.2.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: zeroth-learn
-Version: 0.2.4
+Version: 0.2.5
 Requires-Dist: numpy
 Requires-Dist: pandas
 Requires-Dist: matplotlib

{zeroth_learn-0.2.4 → zeroth_learn-0.2.5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "zeroth-learn"
-version = "0.2.4"
+version = "0.2.5"
 dependencies = [
     "numpy",
     "pandas",

zeroth_learn-0.2.5/zeroth/losses/binary_cross_entropy.py

@@ -0,0 +1,38 @@
+import numpy as np
+
+from ..abstract.loss import Loss
+from ..first_order.layer import Layer
+from ..types import Array
+
+
+class BinaryCrossEntropy(Loss):
+    @staticmethod
+    def compute_loss(Y_pred: Array, Y_true: Array) -> float:
+        Y_true = Y_true.reshape(Y_pred.shape)
+        loss = - (Y_true * np.log(Y_pred + 1e-15) + (1 - Y_true) * np.log(1 - Y_pred + 1e-15))
+        return float(np.mean(loss))
+
+    @staticmethod
+    def compute_batch_losses(Y_pred: Array, Y_true: Array) -> Array:
+        Y_true = Y_true.reshape(Y_pred.shape)
+        loss = -(Y_true * np.log(Y_pred + 1e-15) + (1 - Y_true) * np.log(1 - Y_pred + 1e-15))
+        return np.mean(loss, axis=1)
+
+    @staticmethod
+    def compute_perturbed_losses(pY_pred: Array, Y_true: Array) -> Array:
+        Y_true = Y_true.reshape(1, pY_pred.shape[1], -1)
+        loss = -(Y_true * np.log(pY_pred + 1e-15) + (1 - Y_true) * np.log(1 - pY_pred + 1e-15))
+        return np.mean(loss, axis=2)
+
+    @staticmethod
+    def compute_gradient_wrt_preactivation(last_layer: Layer, Y_pred: Array, Y_true: Array) -> Array:
+        Y_true = Y_true.reshape(Y_pred.shape)
+        return Y_pred - Y_true
+
+    def compute_losses_for_zeroth_order(self, pY_pred: Array, Y_true: Array) -> tuple[float, Array]:
+        avg_loss = self.compute_loss(pY_pred[0], Y_true)
+        p_loss = self.compute_perturbed_losses(pY_pred, Y_true)
+        return avg_loss, p_loss
+
+    def compute_losses_for_first_order(self, last_layer: Layer, Y_pred: Array, Y_true: Array) -> tuple[float, Array]:
+        return self.compute_loss(Y_pred, Y_true), self.compute_gradient_wrt_preactivation(last_layer, Y_pred, Y_true)

{zeroth_learn-0.2.4 → zeroth_learn-0.2.5}/zeroth/utils/activation_functions.py

@@ -17,7 +17,7 @@ class Sigmoid(Activation):
         return 1 / (1 + np.exp(-x))
 
     def derivative(self, x: float | Array) -> float | Array:
-        s = self.__call__(x)  # On réutilise le __call__ pour optimiser
+        s = self.__call__(x)
         return s * (1 - s)
 
 
@@ -42,5 +42,4 @@ class Softmax(Activation):
         return e / np.sum(e, axis=-1, keepdims=True)
 
     def derivative(self, x: float | Array) -> float | Array:
-        # Comme indiqué dans votre code original, pas besoin de dérivée si géré par CrossEntropy
-        raise NotImplementedError("La dérivée de Softmax est généralement combinée avec CrossEntropy.")
+        raise NotImplementedError("no need for derivative when using CrossEntropy loss")

{zeroth_learn-0.2.4 → zeroth_learn-0.2.5}/zeroth/utils/metrics.py

@@ -8,4 +8,4 @@ from ..types import Array
 
 class Accuracy(Metric):
     def __call__(self: Self, Y_pred: Array, Y_true: Array) -> float:
-        return np.sum(np.argmax(Y_pred, axis=1) == Y_true[:, 0]).item() / Y_true.shape[0]
+        return float(np.mean(np.argmax(Y_pred, axis=1) == Y_true[:, 0]))

{zeroth_learn-0.2.4 → zeroth_learn-0.2.5/zeroth_learn.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: zeroth-learn
-Version: 0.2.4
+Version: 0.2.5
 Requires-Dist: numpy
 Requires-Dist: pandas
 Requires-Dist: matplotlib

{zeroth_learn-0.2.4 → zeroth_learn-0.2.5}/zeroth_learn.egg-info/SOURCES.txt

@@ -25,6 +25,7 @@ zeroth/first_order/model.py
 zeroth/first_order/neural_network.py
 zeroth/first_order/optimizers.py
 zeroth/losses/__init__.py
+zeroth/losses/binary_cross_entropy.py
 zeroth/losses/cross_entropy.py
 zeroth/losses/mse.py
 zeroth/utils/__init__.py