ce0325cb72000af404c2c0b4c3ef1644a66bc0e4,examples/mnist.py,,get_new_model,#Any#Any#,76

Before Change

    new_model.tau = model.tau
    train_data = train_data + train_data
    random.shuffle(train_data)
    acc = new_model.train(train_data, batch_size=100)
    return acc, new_model


def sequential_sgd(model, train_data, check_data, n_iter=5):

After Change

def get_new_model(train_data, model, **kwargs):
    kwargs = dict(kwargs)
    learn_rate = np.random.normal(loc=model.eta, scale=0.0001)
    if learn_rate < 0.0001:
        learn_rate = 0.0001

    kwargs["eta"] = learn_rate
    kwargs["mu"] = min(np.random.normal(loc=model.mu, scale=0.01), 0.99)
    kwargs["rho"] = max(0.0, np.random.normal(loc=model.rho, scale=0.0001))
    kwargs["noise"] = max(0.0, np.random.normal(loc=model.noise, scale=0.001))
    kwargs["dropout"] = np.random.normal(loc=model.dropout, scale=0.01)
    new_model = NeuralNet(model.widths, **kwargs)
    new_model.weights = model.weights
    new_model.tau = model.tau
    train_data = train_data + train_data
    random.shuffle(train_data)
    acc = 0.0
    for X, y in train_data:
        acc += new_model.update(Example.dense(model.nr_class, X, y))
    return acc / len(train_data), new_model


def sequential_sgd(model, train_data, check_data, n_iter=5):

Instances