`ptls.nn.head`

All classes from ptls.nn.head also available in ptls.nn

`ptls.nn.Head`

Head is a composition layer. Content is controlled by parameters.

Such scenarios are possible:

Empty layer. Do nothing. Can replace a default head: Head()
L2 norm for output embedding: Head(use_norm_encoder=True)
Binary classification head: Head(objective='classification', input_size=m)
Multiclass classification head: Head(objective='classification', input_size=m, num_classes=n)
Multilayer binary classification head: Head(objective='classification', input_size=m, hidden_layers_sizes=[i, j])
Regression head: Head(objective='regression', input_size=m)

Their combinations are also possible

See docstrings for classes.