def test_predict_regressions(self):
for regressor in ['RandomForestRegressor', 'ExtraTreesRegressor',
'GradientBoostingRegressor', 'AdaBoostRegressor',
'Lasso', 'Ridge', 'ElasticNet',
'KNeighborsRegressor', 'SVR', 'LinearSVR']:
estimator, importances = fit_regressor(
self.table_ecam_fp, self.mdc_ecam_fp, random_state=123,
n_estimators=2, estimator=regressor, n_jobs=1,
missing_samples='ignore')
pred = predict_regression(self.table_ecam_fp, estimator)
exp = self.mdc_ecam_fp.to_series()
# reindex both pred and exp because not all samples present in pred
# are present in the metadata! (hence missing_samples='ignore')
sample_ids = pred.index.intersection(exp.index)
pred = pred.loc[sample_ids]
exp = exp.loc[sample_ids]
# test that expected MSE is achieved (these are mostly quite high
# as we would expect)
mse = mean_squared_error(exp, pred)
# TODO: Remove this conditional when
# https://github.com/qiime2/q2-sample-classifier/issues/193 is
# closed
if regressor == 'Ridge':
self.assertAlmostEqual(
mse, seeded_predict_results[regressor], places=4,
msg='Accuracy of %s regressor was %f, but expected %f' % (
regressor, mse, seeded_predict_results[regressor]))
else:
self.assertAlmostEqual(
mse, seeded_predict_results[regressor],
msg='Accuracy of %s regressor was %f, but expected %f' % (
regressor, mse, seeded_predict_results[regressor]))
def test_predict_feature_order_aint_no_thing(self):
table = self.table_ecam_fp
estimator, importances = fit_regressor(table,
self.mdc_ecam_fp,
random_state=123,
n_estimators=2,
n_jobs=1,
missing_samples='ignore')
# randomly shuffle and reorder features in biom table.
feature_ids = table.ids(axis='observation')
# look ma no seed! we should get the same result no matter the order.
np.random.shuffle(feature_ids)
shuffled_table = table.sort_order(feature_ids, axis='observation')
# now predict values on shuffled data
pred = predict_regression(shuffled_table, estimator)
exp = self.mdc_ecam_fp.to_series()
# reindex both pred and exp because not all samples present in pred
# are present in the metadata! (hence missing_samples='ignore')
sample_ids = pred.index.intersection(exp.index)
pred = pred.loc[sample_ids]
exp = exp.loc[sample_ids]
# test that expected MSE is achieved (these are mostly quite high
# as we would expect)
mse = mean_squared_error(exp, pred)
self.assertAlmostEqual(mse,
seeded_predict_results['RandomForestRegressor'])
def test_fit_regressor(self):
pipeline, importances = fit_regressor(
self.table_ecam_fp, self.mdc_ecam_fp, random_state=123,
n_estimators=2, n_jobs=1, missing_samples='ignore')
exp_imp = pd.read_csv(
self.get_data_path('importance_cv.tsv'), sep='\t', header=0,
index_col=0)
pdt.assert_frame_equal(importances, exp_imp)