chemicalchecker.tool.siamese.siamese_triplets.SiameseTriplets

class SiameseTriplets(model_dir, evaluate=False, predict_only=False, plot=True, save_params=True, generator=None, **kwargs)

Bases: object

Siamese class.

This class implements a simple siamese neural network based on Keras that allows metric learning.

Initialize the Siamese class.

Parameters:

• model_dir (str) – Directorty where models will be stored.

• traintest_file (str) – Path to the traintest file.

• evaluate (bool) – Whether to run evaluation.

Methods

build_model	Compile Keras model
find_lr
fit	Fit the model.
predict	Do predictions.

build_model(input_shape, load=False, cp=None)

Compile Keras model

input_shape(tuple): X dimensions (only nr feat is needed) load(bool): Whether to load the pretrained model.

fit(monitor='val_loss', save=True)

Fit the model.

monitor(str): variable to monitor for early stopping.

predict(x_matrix, dropout_fn=None, dropout_samples=10, cp=False)

Do predictions.

prediction_file(str): Path to input file containing Xs. split(str): which split to predict. batch_size(int): batch size for prediction.