TensorFlow Essentials

1# CPU version
2pip install tensorflow
3
4# GPU version
5pip install tensorflow[and-cuda]
6
7# Check installation
8python -c "import tensorflow as tf; print(tf.__version__)"

 1import tensorflow as tf
 2
 3# Create tensors
 4scalar = tf.constant(42)
 5vector = tf.constant([1, 2, 3])
 6matrix = tf.constant([[1, 2], [3, 4]])
 7
 8# Tensor operations
 9a = tf.constant([[1, 2], [3, 4]])
10b = tf.constant([[5, 6], [7, 8]])
11
12# Element-wise operations
13c = a + b
14d = a * b
15
16# Matrix multiplication
17e = tf.matmul(a, b)
18
19# Reshaping
20reshaped = tf.reshape(a, [4, 1])

 1from tensorflow import keras
 2from tensorflow.keras import layers
 3
 4# Sequential model
 5model = keras.Sequential([
 6    layers.Dense(128, activation='relu', input_shape=(784,)),
 7    layers.Dropout(0.2),
 8    layers.Dense(64, activation='relu'),
 9    layers.Dense(10, activation='softmax')
10])
11
12# Functional API
13inputs = keras.Input(shape=(784,))
14x = layers.Dense(128, activation='relu')(inputs)
15x = layers.Dropout(0.2)(x)
16x = layers.Dense(64, activation='relu')(x)
17outputs = layers.Dense(10, activation='softmax')(x)
18model = keras.Model(inputs=inputs, outputs=outputs)
19
20# Compile
21model.compile(
22    optimizer='adam',
23    loss='sparse_categorical_crossentropy',
24    metrics=['accuracy']
25)

 1# Train model
 2history = model.fit(
 3    x_train, y_train,
 4    batch_size=32,
 5    epochs=10,
 6    validation_split=0.2,
 7    callbacks=[
 8        keras.callbacks.EarlyStopping(patience=3),
 9        keras.callbacks.ModelCheckpoint('best_model.h5', save_best_only=True)
10    ]
11)
12
13# Evaluate
14test_loss, test_acc = model.evaluate(x_test, y_test)
15
16# Predict
17predictions = model.predict(x_new)

 1import tensorflow as tf
 2
 3# Define model, loss, optimizer
 4model = MyModel()
 5loss_fn = keras.losses.SparseCategoricalCrossentropy()
 6optimizer = keras.optimizers.Adam()
 7
 8# Training step
 9@tf.function
10def train_step(x, y):
11    with tf.GradientTape() as tape:
12        predictions = model(x, training=True)
13        loss = loss_fn(y, predictions)
14    
15    gradients = tape.gradient(loss, model.trainable_variables)
16    optimizer.apply_gradients(zip(gradients, model.trainable_variables))
17    return loss
18
19# Training loop
20for epoch in range(epochs):
21    for x_batch, y_batch in dataset:
22        loss = train_step(x_batch, y_batch)

 1# Save entire model
 2model.save('my_model.h5')
 3model.save('my_model')  # SavedModel format
 4
 5# Load model
 6loaded_model = keras.models.load_model('my_model.h5')
 7
 8# Save weights only
 9model.save_weights('weights.h5')
10model.load_weights('weights.h5')

 1# Setup TensorBoard callback
 2tensorboard_callback = keras.callbacks.TensorBoard(
 3    log_dir='./logs',
 4    histogram_freq=1
 5)
 6
 7model.fit(x_train, y_train, callbacks=[tensorboard_callback])
 8
 9# View in terminal
10# tensorboard --logdir=./logs