import numpy as np
import argparse
import os
import shutil
import time
from geocoding import clf_utilities as clf_ut, writers as wrtrs
from geocoding.config import Config
[docs]def main():
"""
Implements the third step of the experiment pipeline. Given a classifier, \
this step is responsible to find the best performing classifier \
configuration.
Returns:
None
"""
# Construct argument parser and parse arguments
ap = argparse.ArgumentParser()
ap.add_argument('-classifier', required=True)
ap.add_argument('-experiment_path', required=True)
args = vars(ap.parse_args())
if not clf_ut.is_valid(args['classifier']):
return
# TODO: add RandomizedSearch except for GridSearch
params_grids = clf_ut.clf_hparams_map[args['classifier']][0]
if isinstance(params_grids, list) is False:
params_grids = [params_grids]
t1 = time.time()
results = []
for i in range(1, Config.n_folds + 1):
fold_path = os.path.join(Config.base_dir, 'experiments', args['experiment_path'], f'features_extraction_results/fold_{i}')
X_train = np.load(os.path.join(fold_path, 'X_train.npy'))
X_test = np.load(os.path.join(fold_path, 'X_test.npy'))
y_train = np.load(os.path.join(fold_path, 'y_train.npy'))
y_test = np.load(os.path.join(fold_path, 'y_test.npy'))
for params_grid in params_grids:
for params in clf_ut.create_clf_params_product_generator(params_grid):
if args['classifier'] == 'XGBoost':
params.update(seed=Config.seed_no)
else: params.update(random_state=Config.seed_no)
clf = clf_ut.clf_callable_map[args['classifier']].set_params(**params)
clf.fit(X_train, y_train)
pred = clf.predict_proba(X_test)
y_pred = np.argsort(-pred, axis=1)[:, :]
info = {'fold': i, 'clf_params': str(params)}
scores = clf_ut.evaluate(y_test, y_pred)
results.append(dict(info, **scores))
results_path = os.path.join(Config.base_dir, 'experiments', args['experiment_path'], 'model_selection_results')
if os.path.exists(results_path):
shutil.rmtree(results_path)
os.makedirs(results_path)
wrtrs.write_clf_space(os.path.join(results_path, 'clf_space.csv'), args['classifier'])
wrtrs.write_results(results_path, results, 'model_selection')
print(f'Model selection done in {time.time() - t1:.3f} sec.')
if __name__ == "__main__":
main()
