Список проголосовавших за комментарий "3Qu, 22:16 не нашёл хорошей книги по нейросетям для scikit. Хайкин — отстал.<br/> Вот пример Deep<br/> <blockquote>model = keras.Sequential ([<br/> layers.Dense (layer_size, activation = «relu»,<br/> kernel_regularizer=regularizers.l2(0.002)),<br/> layers.Dense (layer_size, activation = «relu»,<br/> kernel_regularizer=regularizers.l2(0.002)),<br/> layers.Dense (layer_size, activation = «relu»,<br/> kernel_regularizer=regularizers.l2(0.002)),<br/> layers.Dense (1, activation = «sigmoid»)<br/> ]) # «rmsprop», defaul learning_rate = 0.001<br/> model.compile (optimizer = keras.optimizers.RMSprop(),<br/> loss = «binary_crossentropy»,<br/> metrics = [«accuracy»])<br/> indices_permutation = np.random.permutation (len (train_inputs))<br/> shuffled_inputs = train_inputs [indices_permutation]<br/> shuffled_targets = train_trg_max[indices_permutation]<br/> history = model.fit (shuffled_inputs, shuffled_targets,<br/> epochs = 200, batch_size = 16, validation_split = 0.2)<br/> <br/> </blockquote>это keras"