xdat - pypi Package Compare versions

Comparing version 0.1.295 to 0.1.296

PKG-INFO

 Metadata-Version: 2.4
 Name: xdat
-Version: 0.1.295
+Version: 0.1.296
 Summary: eXtended Data Analysis Toolkit
@@ -5,0 +5,0 @@ Home-page: https://bitbucket.org/hermetric/xdat/

xdat.egg-info/PKG-INFO

 Metadata-Version: 2.4
 Name: xdat
-Version: 0.1.295
+Version: 0.1.296
 Summary: eXtended Data Analysis Toolkit
@@ -5,0 +5,0 @@ Home-page: https://bitbucket.org/hermetric/xdat/

xdat/VERSION

@@ -1,1 +0,1 @@
-0.1.295
+0.1.296

xdat/xkeras.py

@@ -1284,2 +1284,62 @@ import hashlib
 
+class MaxLossCallback(tf.keras.callbacks.Callback):
+    """
+    Stops training when *training* loss dips below a given threshold.
+    Rolls back to the previous epoch's weights (unless it's the first epoch).
+    """
+
+    def __init__(self, threshold, metric="loss", verbose=1):
+        super().__init__()
+        self.threshold = float(threshold)
+        self.metric = metric
+        self.verbose = int(verbose)
+
+        self.prev_weights = None
+        self.prev_loss = None
+        self.prev_epoch = None
+
+    def on_train_begin(self, logs=None):
+        self.prev_weights = None
+        self.prev_loss = None
+        self.prev_epoch = None
+
+    def rollback_to_previous(self):
+        if self.prev_weights is not None:
+            self.model.set_weights(self.prev_weights)
+
+    def on_epoch_end(self, epoch, logs=None):
+        logs = logs or {}
+        current_loss = logs.get(self.metric)
+
+        if current_loss is None:
+            return
+
+        # Stop when loss dips below threshold
+        if float(current_loss) < self.threshold:
+            if self.prev_weights is not None:
+                self.rollback_to_previous()
+                if self.verbose:
+                    print(
+                        f"\nMaxLossCallback: {self.metric}={float(current_loss):.6f} "
+                        f"< {self.threshold:.6f} at epoch={epoch}. "
+                        f"Rolled back to epoch={self.prev_epoch} "
+                        f"({self.metric}={self.prev_loss:.6f}) and stopping."
+                    )
+            else:
+                if self.verbose:
+                    print(
+                        f"\nMaxLossCallback: {self.metric}={float(current_loss):.6f} "
+                        f"< {self.threshold:.6f} at epoch={epoch}. "
+                        f"No previous epoch to rollback to; stopping."
+                    )
+
+            self.model.stop_training = True
+            return
+
+        # Store current epoch as "previous" for potential rollback
+        self.prev_weights = self.model.get_weights()
+        self.prev_loss = float(current_loss)
+        self.prev_epoch = epoch
+
+
 @register_keras_serializable()
@@ -1286,0 +1346,0 @@ def mape_loss(y_true, y_pred):

Alert delta unavailable

Currently unable to show alert delta for PyPI packages.

Improved metrics

Total package byte prevSize

929913

0.23%

Lines of code

9007

0.55%