smash.Net.compile

Net.compile(optimizer='adam', options=None, random_state=None)

Compile the network and set optimizer.

Parameters:

optimizerstr, default ‘adam’

Name of optimizer. Should be one of

• ‘sgd’

• ‘adam’

• ‘adagrad’

• ‘rmsprop’

optionsdict or None, default None

A dictionary of optimizer options.

Hint

See options for each optimizer:

• ‘sgd’ (see here)

• ‘adam’ (see here)

• ‘adagrad’ (see here)

• ‘rmsprop’ (see here)

random_stateint or None, default None

Random seed used to initialize weights.

Note

If not given, the weights will be initialized with a random seed.

Examples

>>> net = smash.Net()

Define graph

>>> net.add(layer="dense", options={"input_shape": (6,), "neurons": 16})
>>> net.add(layer="activation", options={"name": "relu"})

Compile the network

>>> net.compile(optimizer='sgd', options={'learning_rate': 0.009, 'momentum': 0.001})
>>> net
+-------------------------------------------------------+
| Layer Type         Input/Output Shape  Num Parameters |
+-------------------------------------------------------+
| Dense              (6,)/(16,)          112            |
| Activation (ReLU)  (16,)/(16,)         0              |
+-------------------------------------------------------+
Total parameters: 112
Trainable parameters: 112
Optimizer: (sgd, lr=0.009)