Изменения в библиотекте для ЛР2:

- Добавлен callback для частичного отображения эпох
- Добавлены kwargs для частичного отображения эпох и ранней остановки по дельте и значению метрики
- Для EarlyStoppingCallback добавлена зависимость от verbose_show

labworks/LW2/lab02_lib.py

@@ -29,12 +29,14 @@ from pandas import DataFrame
 from sklearn.metrics import precision_score, recall_score, f1_score, confusion_matrix
 from tensorflow.keras.models import Sequential
 from tensorflow.keras.layers import Dense, Activation
+from tensorflow.keras.callbacks import Callback
 
 
 
 visual = True
 verbose_show = False
 
+
 # generate 2d classification dataset
 def datagen(x_c, y_c, n_samples, n_features):
 
@@ -91,8 +93,27 @@ class EarlyStoppingOnValue(tensorflow.keras.callbacks.Callback):
             )
         return monitor_value
 
+
+class VerboseEveryNEpochs(Callback):
+    def __init__(self, every_n_epochs=1000, verbose=1):
+        super().__init__()
+        self.every_n_epochs = every_n_epochs
+        self.verbose = verbose
+
+    def on_epoch_end(self, epoch, logs=None):
+        if (epoch + 1) % self.every_n_epochs == 0:
+            if self.verbose:
+                print(f"\nEpoch {epoch + 1}/{self.params['epochs']}")
+                if logs:
+                    log_str = ", ".join([f"{k}: {v:.4f}" for k, v in logs.items()])
+                    print(f" - {log_str}")
+
+
 #создание и обучение модели автокодировщика
-def create_fit_save_ae(cl_train, ae_file, irefile, epohs, verbose_show, patience):
+def create_fit_save_ae(cl_train, ae_file, irefile, epohs, verbose_show, patience, **kwargs):
+    verbose_every_n_epochs = kwargs.get('verbose_every_n_epochs', 1000)
+    early_stopping_delta = kwargs.get('early_stopping_delta', 0.01)
+    early_stopping_value = kwargs.get('early_stopping_value', 0.0001)
 
     size = cl_train.shape[1]
     #ans = '2'
@@ -140,22 +161,28 @@ def create_fit_save_ae(cl_train, ae_file, irefile, epohs, verbose_show, patience
 
     optimizer = tensorflow.keras.optimizers.Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, amsgrad=False)
     ae.compile(loss='mean_squared_error', optimizer=optimizer)
-    error_stop = 0.0001
     epo = epohs
 
-    early_stopping_callback_on_error = EarlyStoppingOnValue(monitor='loss', baseline=error_stop)
+
+    verbose = 1 if verbose_show else 0
+
+    early_stopping_callback_on_error = EarlyStoppingOnValue(monitor='loss', baseline=early_stopping_value)
     early_stopping_callback_on_improving = tensorflow.keras.callbacks.EarlyStopping(monitor='loss',
-                                                                                           min_delta=0.0001, patience = patience,
-                                                                                           verbose=1, mode='auto',
+                                                                                           min_delta=early_stopping_delta, patience = patience,
+                                                                                           verbose=verbose, mode='min',
                                                                                            baseline=None,
-                                                                                           restore_best_weights=False)
+                                                                                           restore_best_weights=True)
     history_callback = tensorflow.keras.callbacks.History()
-    verbose = 1 if verbose_show else 0
+
     history_object = ae.fit(cl_train, cl_train,
                                 batch_size=cl_train.shape[0],
                                 epochs=epo,
-                                callbacks=[early_stopping_callback_on_error, history_callback,
-                                early_stopping_callback_on_improving],
+                                callbacks=[
+                                  early_stopping_callback_on_error, 
+                                  history_callback,
+                                  early_stopping_callback_on_improving,
+                                  VerboseEveryNEpochs(every_n_epochs=verbose_every_n_epochs),
+                                ],
                                 verbose=verbose)
     ae_trainned = ae
     ae_pred = ae_trainned.predict(cl_train)
@@ -538,4 +565,4 @@ def ire_plot(title, IRE_test, IREth, ae_name):
     plt.gcf().savefig('out/IRE_' + title + ae_name + '.png')
     plt.show()
 
-    return
+    return