def train_test_NN(train, classes, test, **parameters):
parameters['dense2_nonlinearity'] = parameters[
'dense1_nonlinearity'] # hack1
parameters['dense2_init'] = parameters['dense1_init'] # hack2
net = make_net(**parameters)
net.fit(train, classes - 1)
return net.predict_proba(test)
def train_test_NN(train, classes, test, **parameters): train, classes = equalize_class_sizes(train, classes) train, classes = filter_data(train, classes, cut_outlier_frac = 0.06, method = 'OCSVM') # remove ourliers train = normalize_data(train, use_log = True)[0] # also converts to floats test = normalize_data(test, use_log = True)[0] parameters['dense2_nonlinearity'] = parameters['dense1_nonlinearity'] # hack1 parameters['dense2_init'] = parameters['dense1_init'] # hack2 net = make_net(**parameters) net.fit(train, classes - 1) return net.predict_proba(test)
def train_test_NN(train,
labels,
test,
use_rescale_priors=False,
outlier_frac=0,
outlier_method='OCSVM',
normalize_log=True,
use_calibration=False,
**parameters):
net = make_net(**parameters)
train, test = conormalize_data(train, test, use_log=normalize_log)
load_knowledge(net, 'results/nnets/optimize_new.log_1000.net.npz')
prediction = net.predict_proba(test)
if use_rescale_priors:
prediction = scale_to_priors(prediction,
priors=bincount(labels)[1:] /
float64(len(labels)))
return prediction
def train_NN(train,
labels,
test,
outlier_frac=0,
outlier_method='OCSVM',
use_calibration=False,
normalize_log=True,
use_rescale_priors=False,
extra_feature_count=0,
extra_feature_seed=0,
test_data_confidence=None,
test_only=False,
**parameters):
"""
Train a neural network, for internal use by other functions in this file.
"""
train, labels = expand_from_test(train,
labels,
get_testing_data()[0],
confidence=test_data_confidence)
train, test = chain_feature_generators(train,
labels,
test,
extra_features=extra_feature_count,
seed=extra_feature_seed)
train, test = conormalize_data(train, test, use_log=normalize_log)
if outlier_frac:
train, labels = filter_data(train,
labels,
cut_outlier_frac=outlier_frac,
method=outlier_method)
net = make_net(NFEATS=train.shape[1], **parameters)
if use_calibration:
net = CalibratedClassifierCV(net,
method='sigmoid',
cv=ShuffleSplit(train.shape[0],
n_iter=1,
test_size=0.2))
if not test_only:
net.fit(train, labels - 1)
return net, train, test
'dense3_size': None,
'learning_rate': 0.0007, # initial learning rate (learning rate is effectively higher for higher momentum)
'learning_rate_scaling': 100, # progression over time; 0.1 scaled by 10 is 0.01
'momentum': 0.9, # initial momentum
'momentum_scaling': 10 , # 0.9 scaled by 10 is 0.99
'dropout1_rate': 0.5, # [0, 0.5]
'dropout2_rate': None, # inherit dropout1_rate if dense2 exists
'dropout3_rate': None, # inherit dropout2_rate if dense3 exist
'weight_decay': 0.00007, # constrain the weights to avoid overfitting
'max_epochs': 600, # it terminates when overfitting or increasing, so just leave high
'auto_stopping': True, # stop training automatically if it seems to be failing
'save_snapshots_stepsize': None, # save snapshot of the network every X epochs
}
print '>> making network'
net = make_net(train.shape[1], **params)
pretrain = join(PRETRAIN_DIR, '{0:s}_pretrain_{1:d}_{2:d}_{3:d}.net.npz'.format(params['name'], params['dense1_size'] or 0, params['dense2_size'] or 0, params['dense3_size']or 0))
if not isfile(pretrain):
print '>> pretraining network'
make_pretrain(pretrain, train, labels, extra_feature_count = extra_feature_count, **params)
print '>> loading pretrained network'
load_knowledge(net, pretrain)
print '>> training network'
out = net.fit(train, labels - 1)
print '>> saving network'
save_knowledge(net, join(NNET_STATE_DIR, 'single_trained.net.npz'))