from torch.utils.data import random_split
train_ratio, valid_ratio = 0.8, 0.2

# Gesamteanzahl der Trainingsdaten
N_training = len(training_data)

# Berechne die Anzahl der Beispiele für jeden Split
train_size = int(train_ratio * N_training)
valid_size = N_training - train_size

# Teile den Trainingsdatensatz in Train und Valid auf
train_data, valid_data = random_split(training_data, 
                                      [train_size, valid_size])

from torch.utils.data import random_split
train_ratio, valid_ratio = 0.8, 0.2

# Gesamteanzahl der Trainingsdaten
N_training = len(training_data)

# Berechne die Anzahl der Beispiele für jeden Split
train_size = int(train_ratio * N_training)
valid_size = N_training - train_size

# Teile den Trainingsdatensatz in Train und Valid auf
train_data, valid_data = random_split(training_data, 
                                      [train_size, valid_size])

import tensorflow as tf
from sklearn.model_selection import train_test_split
valid_ratio = 0.2

input_data = your_input_data
labels = your_labels

input_train, input_valid, labels_train, labels_valid = 
	train_test_split(input_data, labels, test_size=valid_ratio, random_state=42)

# Create TensorFlow Datasets
train_dataset = tf.data.Dataset.from_tensor_slices((input_train, labels_train))
valid_dataset = tf.data.Dataset.from_tensor_slices((input_valid, labels_valid))

class MLP_var(nn.Module):
    def __init__(self, N_layer: int):
        super(MLP_var, self).__init__()
        layers = []
        layers.append(nn.Linear(28*28, 64))
        layers.append(nn.ReLU())
        for _ in range(N_layer-1):
            layers.append(nn.Linear(64, 64))
            layers.append(nn.ReLU())
        layers.append(nn.Linear(64, 10))
        self.model = nn.Sequential(*layers)

    def forward(self, x):
        x = x.view(x.size(0), -1)
        x = self.model(x)
        return x

for (N_layer, lr) in product(N_layer_values, learning_rate_values):
    model = MLP_var(N_layer)
    model.to(device)
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=lr)

    losses_train, losses_valid, f1_scores_train, f1_scores_valid = full_training(
      model, criterion, optimizer, training_loader, valid_loader, epochs=epochs
    )