ENH: Implemented performance speedup for binary ReliefF + bug fixes

performance_tests.py

@@ -0,0 +1,179 @@
+from skrebate import ReliefF, SURF, SURFstar, MultiSURF, MultiSURFstar
+from sklearn.pipeline import make_pipeline
+from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
+from sklearn.impute import SimpleImputer
+from sklearn.model_selection import cross_val_score
+import pandas as pd
+import numpy as np
+import warnings
+
+import timeit
+
+warnings.filterwarnings('ignore')
+
+np.random.seed(3249083)
+
+genetic_data = pd.read_csv(
+    'data/GAMETES_Epistasis_2-Way_20atts_0.4H_EDM-1_1.tsv.gz', sep='\t', compression='gzip')
+# genetic_data = genetic_data.sample(frac=0.25)
+
+genetic_data_cont_endpoint = pd.read_csv(
+    'data/GAMETES_Epistasis_2-Way_continuous_endpoint_a_20s_1600her_0.4__maf_0.2_EDM-2_01.tsv.gz', sep='\t', compression='gzip')
+genetic_data_cont_endpoint.rename(columns={'Class': 'class'}, inplace=True)
+genetic_data_cont_endpoint = genetic_data_cont_endpoint.sample(frac=0.25)
+
+genetic_data_mixed_attributes = pd.read_csv(
+    'data/GAMETES_Epistasis_2-Way_mixed_attribute_a_20s_1600her_0.4__maf_0.2_EDM-2_01.tsv.gz', sep='\t', compression='gzip')
+genetic_data_mixed_attributes.rename(columns={'Class': 'class'}, inplace=True)
+genetic_data_mixed_attributes = genetic_data_mixed_attributes.sample(frac=0.25)
+
+genetic_data_missing_values = pd.read_csv(
+    'data/GAMETES_Epistasis_2-Way_missing_values_0.1_a_20s_1600her_0.4__maf_0.2_EDM-2_01.tsv.gz', sep='\t', compression='gzip')
+genetic_data_missing_values.rename(columns={'Class': 'class'}, inplace=True)
+genetic_data_missing_values = genetic_data_missing_values.sample(frac=0.25)
+
+genetic_data_multiclass = pd.read_csv('data/3Class_Datasets_Loc_2_01.txt', sep='\t')
+genetic_data_multiclass.rename(columns={'Class': 'class'}, inplace=True)
+genetic_data_multiclass = genetic_data_multiclass.sample(frac=0.25)
+
+
+features, labels = genetic_data.drop('class', axis=1).values, genetic_data['class'].values
+headers = list(genetic_data.drop("class", axis=1))
+
+features_cont_endpoint, labels_cont_endpoint = genetic_data_cont_endpoint.drop(
+    'class', axis=1).values, genetic_data_cont_endpoint['class'].values
+headers_cont_endpoint = list(genetic_data_cont_endpoint.drop("class", axis=1))
+
+features_mixed_attributes, labels_mixed_attributes = genetic_data_mixed_attributes.drop(
+    'class', axis=1).values, genetic_data_mixed_attributes['class'].values
+headers_mixed_attributes = list(genetic_data_mixed_attributes.drop("class", axis=1))
+
+features_missing_values, labels_missing_values = genetic_data_missing_values.drop(
+    'class', axis=1).values, genetic_data_missing_values['class'].values
+headers_missing_values = list(genetic_data_missing_values.drop("class", axis=1))
+
+features_multiclass, labels_multiclass = genetic_data_multiclass.drop(
+    'class', axis=1).values, genetic_data_multiclass['class'].values
+headers_multiclass = list(genetic_data_multiclass.drop("class", axis=1))
+
+
+
+# Basic Parallelization Tests and Core binary data and discrete feature data testing (Focus on ReliefF only for efficiency)------------------------------------------------------------
+def test_relieff():
+    """Check: Data (Binary Endpoint, Discrete Features): ReliefF works in a sklearn pipeline"""
+    np.random.seed(49082)
+
+    alg = ReliefF(n_features_to_select=2, n_neighbors=10)
+    alg.fit(features, labels)
+
+
+def test_relieff_parallel():
+    """Check: Data (Binary Endpoint, Discrete Features): ReliefF works in a sklearn pipeline when ReliefF is parallelized"""
+    # Note that the rebate algorithm cannot be parallelized with both the random forest and the cross validation all at once.  If the rebate algorithm is parallelized, the cross-validation scoring cannot be.
+    np.random.seed(49082)
+
+    alg = ReliefF(n_features_to_select=2, n_neighbors=10, n_jobs=-1)
+    alg.fit(features, labels)
+
+
+def test_relieffpercent():
+    """Check: Data (Binary Endpoint, Discrete Features): ReliefF with % neighbors works in a sklearn pipeline"""
+    np.random.seed(49082)
+
+    alg = ReliefF(n_features_to_select=2, n_neighbors=0.1)
+    alg.fit(features, labels)
+
+
+def test_surf():
+    """Check: Data (Binary Endpoint, Discrete Features): SURF works in a sklearn pipeline"""
+    np.random.seed(240932)
+
+    alg = SURF(n_features_to_select=2)
+    alg.fit(features, labels)
+
+
+def test_surf_parallel():
+    """Check: Data (Binary Endpoint, Discrete Features): SURF works in a sklearn pipeline when SURF is parallelized"""
+    np.random.seed(240932)
+
+    alg = SURF(n_features_to_select=2, n_jobs=-1)
+    alg.fit(features, labels)
+
+
+def test_surfstar():
+    """Check: Data (Binary Endpoint, Discrete Features): SURF* works in a sklearn pipelined"""
+    np.random.seed(9238745)
+
+    alg = SURFstar(n_features_to_select=2)
+    alg.fit(features, labels)
+
+
+def test_surfstar_parallel():
+    """Check: Data (Binary Endpoint, Discrete Features): SURF* works in a sklearn pipeline when SURF* is parallelized"""
+    np.random.seed(9238745)
+
+    alg = SURFstar(n_features_to_select=2, n_jobs=-1)
+    alg.fit(features, labels)
+
+
+def test_multisurfstar():
+    """Check: Data (Binary Endpoint, Discrete Features): MultiSURF* works in a sklearn pipeline"""
+    np.random.seed(320931)
+
+    alg = MultiSURFstar(n_features_to_select=2)
+    alg.fit(features, labels)
+
+
+def test_multisurfstar_parallel():
+    """Check: Data (Binary Endpoint, Discrete Features): MultiSURF* works in a sklearn pipeline when MultiSURF* is parallelized"""
+    np.random.seed(320931)
+
+    alg = MultiSURFstar(n_features_to_select=2, n_jobs=-1)
+    alg.fit(features, labels)
+
+
+def test_multisurf():
+    """Check: Data (Binary Endpoint, Discrete Features): MultiSURF works in a sklearn pipeline"""
+    np.random.seed(320931)
+
+    alg = MultiSURF(n_features_to_select=2)
+    alg.fit(features, labels)
+
+
+def test_multisurf_parallel():
+    """Check: Data (Binary Endpoint, Discrete Features): MultiSURF works in a sklearn pipeline when MultiSURF is parallelized"""
+    np.random.seed(320931)
+
+    alg = MultiSURF(n_features_to_select=2, n_jobs=-1)
+    alg.fit(features, labels)
+
+
+test_cases = [
+    test_relieff,
+    test_relieff_parallel,
+    test_relieffpercent,
+    # test_surf,
+    # test_surf_parallel,
+    # test_surfstar,
+    # test_surfstar_parallel,
+    # test_multisurfstar,
+    # test_multisurfstar_parallel,
+    # test_multisurf,
+    # test_multisurf_parallel
+]
+
+if __name__ == '__main__':
+    timing_df = pd.DataFrame(columns=['test_case', 'mean', 'std'])
+
+    for test_case in test_cases:
+        timing = timeit.repeat(test_case, number=1, repeat=5)
+        # ignore the first test to avoid high initial overhead to compile numba
+        # functions with small datasets
+        timing = timing[1:]
+        print(test_case.__name__, np.mean(timing), np.std(timing))
+        d = {'test_case' : test_case.__name__, 'mean' : np.mean(timing), 'std' : np.std(timing)}
+        timing_df = timing_df.append(d, ignore_index = True)
+
+    print(timing_df)
+
+    timing_df.to_csv('timing_benchmarks.csv')

run_performance_benchmark.sh

@@ -0,0 +1,2 @@
+python -m cProfile -o perf_data.pstats performance_tests.py
+gprof2dot -f pstats perf_data.pstats | dot -Tpng -o perfgraph.png