# self.meta_model_.fit(out_of_fold_predictions, y) # return self # # #Do the predictions of all base models on the test data and use the averaged predictions as # #meta-features for the final prediction which is done by the meta-model # def predict(self, X): # meta_features = np.column_stack([ # np.column_stack([model.predict(X) for model in base_models]).mean(axis=1) # for base_models in self.base_models_ ]) # return self.meta_model_.predict(meta_features) # In[1]: from models.cnn import CNN model = CNN() model.fit(X_train[0], y_train[0], X_test[0], y_test[0]) # # Make submission # In[11]: #import pandas as pd #sub = pd.DataFrame() # In[ ]:
def train(train_ds, val_ds, exp_count, overwrite_flag, **params):
if overwrite_flag:
tag = exp_count
else:
tag_list = [
int(path.split("_")[-1]) for path in os.listdir("results")
if "exp" in path
]
if tag_list:
tag = max(tag_list) + 1
else:
tag = 0
save_dir = 'experiment_' + str(tag)
save_dir = os.path.join('results', save_dir)
if not os.path.isdir(save_dir):
os.makedirs(save_dir)
num_epochs = params["num_epochs"]
learning_rate = params["learning_rate"]
optimizer_type = params["optimizer_type"]
initializer_type = params["initializer_type"]
loss_type = params["loss_type"]
early_stop_tolerance = params["early_stop_tolerance"]
optimizer_dispatcher = {
"adam": tf.optimizers.Adam,
"sgd": tf.optimizers.SGD,
"rmsprop": tf.optimizers.RMSprop
}
initializer_dispatcher = {
"xavier": tf.initializers.glorot_uniform,
"random": tf.random_normal_initializer
}
loss_dispatcher = {"l2": loss_l2, "l1": loss_l1}
optimizer = optimizer_dispatcher[optimizer_type](learning_rate)
initializer = initializer_dispatcher[initializer_type]()
loss_fun = loss_dispatcher[loss_type]
model = CNN(initializer=initializer,
optimizer=optimizer,
loss_function=loss_fun,
**params)
train_loss, val_loss, evaluation_val_loss = model.fit(
train_ds, val_ds, num_epochs, early_stop_tolerance)
# plot and save the loss curve
plot_loss_curve(train_loss, val_loss, loss_type, save_dir)
save_dict = params
save_dict['train_loss'] = train_loss
save_dict['val_loss'] = val_loss
save_dict['evaluation_val_loss'] = evaluation_val_loss
print('Saving...')
conf_save_path = os.path.join(save_dir, 'config.pkl')
model_save_path = os.path.join(save_dir, 'model.pkl')
for path, obj in zip([conf_save_path, model_save_path],
[save_dict, model]):
with open(path, 'wb') as file:
pkl.dump(obj, file)
