Galaxy |

Build Deep learning Batch Training Models (version 1.0.11.0)

Choose a building mode:

Select the dataset containing model configurations (JSON):

Select a learning class:

Select a batch data generator:

seq_length:

Integer. Sequence length or number of bases.

shuffle:

Whether to shuffle the data between epochs.

seed:

Integer, random seed for data shuffling

Compile Parameters

Compile Parameters 0

Fit Parameters

Fit Parameters 0

class_positive_factor:

For binary classification only. If int, like 5, will convert to class_weight {0: 1, 1: 5}. If float, 0.2, corresponds to class_weight {0: 1/0.2, 1: 1}

prediction_steps:

Prediction steps. Optional. If None, it equals number of samples divided by `batch_size`.

Random Seed:

Integer or blank for None. Warning: when random seed is set to an integer, training will be running in single thread mode, which may cause slowness.

What does this tool do?

This tool builds deep learning training models using API galaxy_ml.keras_galaxy_model.KerasGBatchClassifier, which takes parameters in FIVE categories.

a JSON file that contains layer information for a deep learning model.
a data batch generator that converts raw data, such as images and genomic sequences, into numerical data to be able to fit the deep learning model. That the cycle of batch conversion - fitting occur in stream mode, also called on-line transformation, guarantees the training to be CPU and memory efficient. Reference: galaxy_ml.preprocessors.FastaDNABatchGenerator, galaxy_ml.preprocessors.FastaRNABatchGenerator, galaxy_ml.preprocessors.FastaProteinBatchGenerator, galaxy_ml.preprocessors.GenomicIntervalBatchGenerator.
compile parameters, are mainly composed of loss function and optimizer.
fit parameters, a group of variables that control the training process, referring to galaxy_ml.keras_galaxy_model.KerasGBatchClassifier and Keras.
other parameters, including class_positive_factor, prediction_steps, seed (random seed) and so on.

Output

A model file that could be used in model_validation tool or hyperparameter search tool.