simclr

SimCLR Trainer API

This module contains the implementation of a SimCLR trainer that can be used to train a model using the SimCLR approach.

Classes

Name	Description
SimCLRTrainer	A trainer for SimCLR

Machine-readable model

• SimCLRTrainerclassCreates a SimCLRTrainer.

class

SimCLRTrainer

Python

Creates a SimCLRTrainer.

helia_edge/trainers/simclr.py:17

SimCLRTrainer(encoder: keras.Model, projector: keras.Model | None = None, **kwargs={})

Creates a SimCLRTrainer.

If no projector is provided, a default one will be created based on paper.

Base class: ContrastiveTrainer.

Inherited from ContrastiveTrainer: call(), fit(), linear_probe(), run_augmenters(), save(), test_step(), train_step().

Parameters

Parameters of SimCLRTrainer

Name	Type	Default	Description
encoder	keras.Model	Required	The encoder model.
projector	keras.Model	None	The projector model. Defaults to None.

method

compile

Python

helia_edge/trainers/simclr.py:53

compile(
    encoder_optimizer: keras.Optimizer,
    encoder_loss: keras.Loss | None = None,
    encoder_metrics: list[keras.Metric] | None = None,
    probe_optimizer: keras.Optimizer | None = None,
    probe_loss: keras.Loss | None = None,
    probe_metrics: list[keras.Metric] | None = None,
    **kwargs={},
)