autoencoder.py 文件源码

def fit(self, 
            data_x_train,
            data_x_dev=None,
            data_x_test=None,
            n_epochs=10,
            batch_size=10):
        assert n_epochs > 0
        assert batch_size < data_x_train.shape[0]

        size_x_train = data_x_train.shape[0]

        n_batches = size_x_train / batch_size

        for e in range(n_epochs):
            epoch_costs = np.zeros(n_batches)
            bar = tqdm(range(n_batches), desc='Epoch: {:d}'.format(e))

            for i in bar:
                batch_x = data_x_train[i*batch_size:(i+1)*batch_size]
                err = self.partial_fit(batch_x)
                epoch_costs[i] = err

            mean_cost = epoch_costs.mean()
            print 'Train error: {:.4f}'.format(mean_cost)

            if data_x_dev is not None:
                random_indices = np.random.randint(0, data_x_dev.shape[0], batch_size)
                batch_x = data_x_dev[random_indices]
                err = self.get_cost(batch_x)
                print 'Validation data error: {:.4f}'.format(err)

        if data_x_test is not None:
                err = self.get_cost(data_x_test)
                print 'Test data error: {:.4f}'.format(err)