import numpy as np
from sklearn import datasets
from sklearn.cross_validation import train_test_split
from chainer import cuda
import skchainer as skc
# cuda.init()
digits = datasets.load_digits()
X = digits.data.astype(np.float32)
y = digits.target.astype(np.int32)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=0)
tuned_parameters = [
{'opt_type': ['sgd'], 'opt_lr': [0.01, 0.001], 'net_hidden': [50, 100, 200, 300]},
{'opt_type': ['adagrad'], 'opt_lr': [0.01, 0.001], 'net_hidden': [50, 100, 200, 300]},
{'opt_type': ['adam'], 'net_hidden': [50, 100, 200, 300]}
]
model = skc.LogisticRegressionEstimator(epochs=50, net_out=10, threshold=1e-6)
skc.grid_search(model, tuned_parameters, X_train, y_train, X_test, y_test, score='accuracy')
digits = datasets.load_digits()
X = digits.data.astype(np.float32)
y = digits.target.astype(np.int32)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.3,
random_state=0)
tuned_parameters = [{
'opt_type': ['sgd'],
'opt_lr': [0.01, 0.001],
'net_hidden': [50, 100, 200, 300]
}, {
'opt_type': ['adagrad'],
'opt_lr': [0.01, 0.001],
'net_hidden': [50, 100, 200, 300]
}, {
'opt_type': ['adam'],
'net_hidden': [50, 100, 200, 300]
}]
model = skc.LogisticRegressionEstimator(epochs=50, net_out=10, threshold=1e-6)
skc.grid_search(model,
tuned_parameters,
X_train,
y_train,
X_test,
y_test,
score='accuracy')
parser.add_argument('--n_jobs', type=int, default=-1)
parser.add_argument('--n_iter', type=int, default=10)
parser.add_argument('--gpu', type=int, default=-1)
parser.add_argument('--epochs', type=int, default=1)
args = parser.parse_args()
model = skc.RNNCharEstimator(epochs=args.epochs, vocab_size=len(vocab), threshold=1e-6)
if args.mode == 'grid':
tuned_parameters = [
{'net_type': ['irnn'],
'opt_type': ['adam'],
'opt_lr': [0.01],
'net_hidden': [200],
'batch_size': [3, 4, 5, 6, 8, 10, 15],
'gpu': [args.gpu]}
]
skc.grid_search(model, tuned_parameters, X_train, y_train, X_test, y_test, score='accuracy', n_jobs=args.n_jobs)
elif args.mode == 'random':
tuned_parameters = {
'net_type': ['irnn'], 'opt_type': ['adam', 'adagrad'], 'net_hidden': sp.stats.norm(300, 100)
}
skc.random_search(model, tuned_parameters, X_train, y_train, X_test, y_test, score='accuracy', n_jobs=args.n_jobs, n_iter=args.n_iter)
elif args.mode == 'time':
print("hello")
# predict -----------------------
# print model.predict(X[1:5,])