LIMIT = 9000
DICTIONARY_SIZE = 600
sofia_path = None # sys.argv[1]
if len(sys.argv) > 1:
VOX_FEATURES = sys.argv[1] == "use_voxforge"
else:
VOX_FEATURES = False
print "Using Vox features", VOX_FEATURES
print "LIMIT", LIMIT
print "Checking class names"
_, REVERSE_CLASSES = load_classes_info()
print "Loading train"
X_train_transformed = load_train_features("bow_train_features.pcl", extract_mfcc_features, limit=LIMIT)
X_train = np.vstack([f for (_, _, f) in X_train_transformed if f is not None])
# X_train = np.reshape(X_train, ( len(X_train_transformed), num_features ) )
print X_train.shape
# for filename, lbl, features in X_train_transformed:
# if features is None:
# print "Train",filename, "is none"
# continue
# if X_total is not None:
# X_total = np.vstack([ X_total, features ])
# else:
# X_total = features
X_total = X_train
print "Ready with train"
print "Starting train"
test_data = get_all_test_data()
from sklearn.grid_search import GridSearchCV
from sknn.mlp import Classifier, Layer
from sklearn.ensemble import RandomForestClassifier,GradientBoostingClassifier
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC
__author__ = 'egor'
VALIDATE = True
np.random.seed(100500)
VOX_FEATURES = True
LIMIT = 5000
print "Using Vox features", VOX_FEATURES
print "LIMIT", LIMIT
X_train_transformed = load_train_features('gmm_train_features.pcl', extract_gmm_feature, limit = LIMIT )
print "Loaded train"
test_data = get_all_test_data()
filename_by_path = {path : filename for (path, filename) in test_data }
X_test_transformed = load_test_features('gmm_test_features.pcl', extract_gmm_feature, limit = LIMIT)
print "Loaded test"
if VOX_FEATURES:
print "Loading voxforge features"
if LIMIT is None:
vox_limit = 4000
else:
vox_limit = LIMIT
voxforge_features = load_vox_forge_files('/store/egor/voxforge', vox_limit, 'gmm_voxfeatures.pcl', extract_gmm_feature )
else:
