blob: 94464fe3632249b76e3b4d52240f79a95822b69d (
plain) (
tree)
|
|
from blocks.initialization import IsotropicGaussian, Constant
import data
from model.dest_mlp import Model, Stream
n_begin_end_pts = 5 # how many points we consider at the beginning and end of the known trajectory
dim_embeddings = [] # do not use embeddings
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
max_splits = 100
|
