from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.datasets import mnist
def get_mnist_model():
inputs = keras.Input(shape=(28*28,))
features = layers.Dense(512, activation="relu")(inputs)
features = layers.Dropout(0.5)(features)
outputs = layers.Dense(10, activation="softmax")(features)
model = keras.Model(inputs, outputs)
return model
(images, labels), (test_images, test_labels) = mnist.load_data()
images = images.reshape((60000, 28*28)).astype("float32") / 255
test_images = test_images.reshape((10000, 28*28)).astype("float32") / 255
train_images, val_images = images[10000:], images[:10000]
train_labels, val_labels = labels[10000:], labels[:10000]
model = get_mnist_model()
model.compile(optimizer="rmsprop",
loss="sparse_categorical_crossentropy",
metrics=["accuracy"])
tensorboard = keras.callbacks.TensorBoard(
log_dir="/full_path_to_your_log_dir",
)
model.fit(train_images, train_labels,
epochs=10,
callbacks=[tensorboard],
validation_data=(val_images, val_labels))
<aside> 🖥️ %load_ext tensorboard %tensorboard --logdir /full_path_to_your_log_dir
</aside>