Olá, durante todo o curso, eu notei que toda vez que treinei minha rede, ela não processava as 48mil imagens, e sim 1500 quando eu não utilizava
batch_size, e ainda menos imagens ao utilizar o batch_size. Quando eu rodo o shape de imagens_treino, existem 60mil imagens: (60000, 28, 28).
Mas quando rodo o treinamento do modelo:
modelo = keras.Sequential([
keras.layers.Flatten(input_shape=(28,28)),
keras.layers.Dense(256,activation=tensorflow.nn.relu),
keras.layers.Dropout(0.2),
#keras.layers.Dense(128,activation=tensorflow.nn.relu),
#keras.layers.Dense(64,activation=tensorflow.nn.relu),
keras.layers.Dense(10,activation=tensorflow.nn.softmax),
])
adam = keras.optimizers.Adam(learning_rate=0.002)
parando_cedo = [keras.callbacks.EarlyStopping(monitor='val_loss')]
modelo.compile(optimizer=adam , loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
historico = modelo.fit(imagens_treino,identificacoes_treino, batch_size=150 ,epochs=5, validation_split = 0.2 , callbacks=parando_cedo)
Só são processadas pouquíssimas imagens:
Epoch 1/5
320/320 ━━━━━━━━━━━━━━━━━━━━ 7s 18ms/step - accuracy: 0.7345 - loss: 0.7503 - val_accuracy: 0.8499 - val_loss: 0.4100
Epoch 2/5
320/320 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - accuracy: 0.8523 - loss: 0.4144 - val_accuracy: 0.8638 - val_loss: 0.3750
Epoch 3/5
320/320 ━━━━━━━━━━━━━━━━━━━━ 5s 8ms/step - accuracy: 0.8660 - loss: 0.3663 - val_accuracy: 0.8711 - val_loss: 0.3576
Epoch 4/5
320/320 ━━━━━━━━━━━━━━━━━━━━ 6s 11ms/step - accuracy: 0.8758 - loss: 0.3400 - val_accuracy: 0.8751 - val_loss: 0.3447
Epoch 5/5
320/320 ━━━━━━━━━━━━━━━━━━━━ 3s 8ms/step - accuracy: 0.8807 - loss: 0.3194 - val_accuracy: 0.8736 - val_loss: 0.3411
Alguém pode me ajudar a entender?
Abraço
