blob: accb611013df9c63c08d8ca3217b0742881a9c38 (
plain) (
tree)
|
|
from blocks.initialization import IsotropicGaussian, Constant
import model.dest_simple_mlp as model
import data
n_begin_end_pts = 5 # how many points we consider at the beginning and end of the known trajectory
n_end_pts = 5
n_valid = 1000
dim_embeddings = [
('origin_call', data.origin_call_train_size, 10),
('origin_stand', data.stands_size, 10)
]
dim_input = n_begin_end_pts * 2 * 2 + sum(x for (_, _, x) in dim_embeddings)
dim_hidden = [200, 100]
dim_output = 2
embed_weights_init = IsotropicGaussian(0.001)
mlp_weights_init = IsotropicGaussian(0.01)
mlp_biases_init = Constant(0.001)
learning_rate = 0.0001
momentum = 0.99
batch_size = 32
valid_set = 'cuts/test_times_0'
|
