自定义回调以保存少量模型

问：

我正在开发新的深度学习模型，我需要创建自定义回调，该回调将 GridSearch 中每个 hiperparameters 组合的最佳权重保存到单独的文件中。例如，当它在训练结束时用 batch_size=16 和 activation='relu' 检查模型时，它应该将最佳权重保存到文件batch_16_activation_relu_hdf5。然后batch_32_activation_tanh_hdf5等。理想情况下，它还应该在 txt 文件中添加新行，其中包含这些信息以及val_accuracy在训练期间获得的信息。这就是我完成的内容：

import os
import numpy as np
from keras.models import Sequential
from keras.layers import Dense
from keras.datasets import mnist
from keras.utils import to_categorical
from scikeras.wrappers import KerasClassifier
from sklearn.model_selection import GridSearchCV
from keras.callbacks import Callback

# Definiuj customowy callback do zapisywania najlepszych wag
class SaveBestModelPerIteration(Callback):
    def __init__(self):
        super(SaveBestModelPerIteration, self).__init__()
        self.best_val_loss = float('inf')

    def on_epoch_end(self, logs=None):
        current_val_loss = logs.get('val_loss')
        if current_val_loss is not None and current_val_loss < self.best_val_loss:
            self.best_val_loss = current_val_loss
            current_val_accuracy = logs.get('val_accuracy')
            optimizer = self.params['optimizer']
            activation = self.params['activation']
            batch_size = self.params['batch_size']
            model_filename = f"optimizer_{optimizer}_activation_{activation}_batch_size_{batch_size}_val_loss_{current_val_loss:.4f}.hdf5"
            self.model.save(model_filename)
            print(f"Model saved as {model_filename} with validation loss: {current_val_loss:.4f}")
            with open('model.txt', 'w') as f:
                f.write(f"Optimizer: {optimizer}, Activation: {activation}, Batch Size: {batch_size}, Val Loss: {current_val_loss:.4f}, Val Accuracy: {current_val_accuracy:.4f}\n")
            
# Załaduj dane do klasyfikacji
(X_train, y_train), (X_test, y_test) = mnist.load_data()
X_train = X_train.reshape(X_train.shape[0], 784).astype('float32') / 255
X_test = X_test.reshape(X_test.shape[0], 784).astype('float32') / 255
y_train = to_categorical(y_train, 10)
y_test = to_categorical(y_test, 10)

# Definiuj funkcję tworzącą model
def create_model(neurons, activation, optimizer):
    model = Sequential()
    model.add(Dense(neurons, input_dim=784, activation=activation))
    model.add(Dense(10, activation='softmax'))
    model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])
    return model

# Utwórz model KerasClassifier
model = KerasClassifier(model=create_model, epochs=10, batch_size=128)

# Definiuj przestrzeń parametrów do przeszukiwań
params={
'model__neurons':[32,64,128,256],
'model__activation':['relu', 'tanh', 'sigmoid', 'linear'],
'model__optimizer':['adam','rmsprop'],
}

save_weights_callback = SaveBestModelPerIteration()

# Utwórz obiekt przeszukiwań siatki
grid = GridSearchCV(estimator=model, param_grid=params, n_jobs=1, cv=3)

# Wykonaj przeszukiwania siatki
grid_result = grid.fit(X_train, y_train, callbacks=[save_weights_callback])

# Wydrukuj najlepsze wyniki przeszukiwań
print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))

不幸的是，我很难做到这一点，我没有任何其他想法。