Massive memory usage running the LazyClassifier

[qemtek]

Describe the bug Using a dataset with 500k rows and 27 features, I ran into a huge memory issue on iteration 12/30. Screenshot included so you can see how much memory was being used.

Desktop (please complete the following information):

• OS: OSX Catalina 10.15.5

Additional context Other packages installed

awswrangler==2.4.0 pandas==1.2.1 numpy==1.20.0 scikit-learn==0.23.1 sqlalchemy==1.3.23 psycopg2-binary==2.8.6 lazypredict==0.2.7 tqdm==4.56.0 xgboost==1.3.3 lightgbm==3.1.1 pytest==6.2.2 imblearn shap==0.38.1 matplotlib==3.3.4 ipython

[apostolides]

Hello,

I have the same issue using a train dataset with 125K rows. I'm training the models on google colaboratory with12G ram available. Runtime crashes on 38% prompting a huge amount of allocated memory. Did you find any workarounds for this issue?

Thanks in advance.

[felixvor]

A workaround is to filter out high memory model architectures from the default regressors / classifiers list and to pass that custom list of models to the LazyRegressor / LazyClassifier. For example:

import lazypredict
from lazypredict.Supervised import LazyRegressor

highmem_regressors = [
    "GammaRegressor", "GaussianProcessRegressor", "KernelRidge", "QuantileRegressor"
]
regressors = [reg for reg in lazypredict.Supervised.REGRESSORS if reg[0] not in highmem_regressors]
reg = LazyRegressor(regressors=regressors, verbose=1, ignore_warnings=True, custom_metric=None)
models, predictions = reg.fit(X_train, X_test, y_train, y_test)

[dvijkalsi]

This worked for me, I was using Google collab 8GB RAM

highmem_classifiers = ["LabelSpreading","LabelPropagation","BernoulliNB","KNeighborsClassifier", "ElasticNetClassifier", "GradientBoostingClassifier", "HistGradientBoostingClassifier"]

# Remove the high memory classifiers from the list
classifiers = [c for c in lazypredict.Supervised.CLASSIFIERS if c[0] not in highmem_classifiers]

clf = LazyClassifier(classifiers=classifiers, verbose=1, ignore_warnings=True, custom_metric=None)
models, predictions = clf.fit(X_train, X_test, y_train, y_test)
model_dictionary = clf.provide_models(X_train, X_test, y_train, y_test)
models