Galaxy-ML

Galaxy-ML is a web machine learning end-to-end pipeline building framework, with special support to biomedical data. Under the management of unified scikit-learn APIs, cutting-edge machine learning libraries are combined together to provide thousands of different pipelines suitable for various needs. In the form of Galalxy tools, Galaxy-ML provides scalabe, reproducible and transparent machine learning computations.

Key features

  • simple web UI
  • no coding or minimum coding requirement
  • fast model deployment and model selection, specialized in hyperparameter tuning using GridSearchCV
  • high level of parallel and automated computation

Supported modules

A typic machine learning pipeline is composed of a main estimator/model and optional preprocessing component(s).

Model
  • scikit-learn
    • sklearn.ensemble
    • sklearn.linear_model
    • sklearn.naive_bayes
    • sklearn.neighbors
    • sklearn.svm
    • sklearn.tree
  • xgboost
    • XGBClassifier
    • XGBRegressor

  • mlxtend

    • StackingCVClassifier
    • StackingClassifier
    • StackingCVRegressor
    • StackingRegressor

  • Keras (Deep learning models are re-implemented to fully support sklearn APIs. Supports parameter, including layer subparameter, swaps or searches. Supports callbacks)

    • KerasGClassifier
    • KerasGRegressor
    • KerasGBatchClassifier (works best with online data generators, processing images, genomic sequences and so on)

  • BinarizeTargetClassifier/BinarizeTargetRegressor

  • IRAPSClassifier
Preprocessor
  • scikit-learn
    • sklearn.preprocessing
    • sklearn.feature_selection
    • sklearn.decomposition
    • sklearn.kernel_approximation
    • sklearn.cluster
  • imblanced-learn
    • imblearn.under_sampling
    • imblearn.over_sampling
    • imblearn.combine
  • skrebate
    • ReliefF
    • SURF
    • SURFstar
    • MultiSURF
    • MultiSURFstar
  • TDMScaler
  • DyRFE/DyRFECV
  • Z_RandomOverSampler
  • GenomeOneHotEncoder
  • ProteinOneHotEncoder
  • FastaDNABatchGenerator
  • FastaRNABatchGenerator
  • FastaProteinBatchGenerator
  • GenomicIntervalBatchGenerator
  • GenomicVariantBatchGenerator
  • ImageDataFrameBatchGenerator

Installation

APIs for models, preprocessors and utils implemented in Galaxy-ML can be installed separately.

conda install -c bioconda -c conda-forge Galaxy-ML
Installing using pip
pip install -U Galaxy-ML
Installing from source
python setup.py install
Using source code inplace
python setup.py build_ext --inplace

To install Galaxy-ML tools in Galaxy, please refer to https://galaxyproject.org/admin/tools/add-tool-from-toolshed-tutorial/.

Examples for using Galaxy-ML custom models

# handle imports
from keras.models import Sequential
from keras.layers import Dense, Activation
from sklearn.model_selection import GridSearchCV
from galaxy_ml.keras_galaxy_models import KerasGClassifier


# build a DNN classifier
model = Sequential()
model.add(Dense(64))
model.add(Activation(‘relu'))
model.add((Dense(1, activation=‘sigmoid’)))
config = model.get_config()

classifier = KerasGClassifier(config, random_state=42)


# clone a classifier
clf = clone(classifier)


# Get parameters
params = clf.get_params()


# Set parameters
new_params = dict(
    epochs=60,
    lr=0.01,
    layers_1_Dense__config__kernel_initializer__config__seed=999,
    layers_0_Dense__config__kernel_initializer__config__seed=999
)
clf.set_params(**new_params)


# model evaluation using GridSearchCV
grid = GridSearchCV(clf, param_grid={}, scoring=‘roc_auc’, cv=5, n_jobs=2)
grid.fit(X, y)

Example for using Galaxy-ML to persist a sklearn/keras model

from galaxy_ml.model_persist import (dump_model_to_h5,
                                     load_model_from_h5)

# dump model to hdf5
dump_model_to_h5(model, `save_path`,
                 store_hyperparameter=True)

# load model from hdf5
model = load_model_from_h5(`path_to_hdf5`)

Performance comparison

Galaxy-ML's HDF5 saving utils perform faster than cPickle for large, array-rich models.

Loading model using pickle...
(1.2471628189086914 s)

Dumping model using pickle...
(3.6942389011383057 s)
File size: 930712861

Dumping model to hdf5...
(3.006715774536133 s)
File size: 930729696

Loading model from hdf5...
(0.6420958042144775 s)

Pipeline(memory=None,
         steps=[('robustscaler',
                 RobustScaler(copy=True, quantile_range=(25.0, 75.0),
                              with_centering=True, with_scaling=True)),
                ('kneighborsclassifier',
                 KNeighborsClassifier(algorithm='auto', leaf_size=30,
                                      metric='minkowski', metric_params=None,
                                      n_jobs=1, n_neighbors=100, p=2,
                                      weights='uniform'))],
         verbose=False)