TensorFlow & Keras · Module 4 of 10

Compiling and training

Train models with compile and fit.

📖 Key ideas

model.compile(optimizer="adam", loss="sparse_categorical_crossentropy", metrics=["accuracy"]).
model.fit(x, y, epochs=10, batch_size=32, validation_split=0.2).
History object stores loss and metrics per epoch.
model.evaluate() tests on held-out data.
model.predict() makes predictions.

💡 Real-life example: compile chooses the training plan; fit runs the training sessions; evaluate is the final exam.

compile()Sets optimiser, loss, metrics

fit()Trains the model

evaluate()Measures performance on data

predict()Generates predictions