from blocks.initialization import IsotropicGaussian, Constant
import data
from model.time_simple_mlp import Model, Stream
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.n_train_clients+1, 10),
('origin_stand', data.n_stands+1, 10),
('week_of_year', 52, 10),
('day_of_week', 7, 10),
('qhour_of_day', 24 * 4, 10),
('day_type', 3, 10),
('taxi_id', 448, 10),
]
dim_input = n_begin_end_pts * 2 * 2 + sum(x for (_, _, x) in dim_embeddings)
dim_hidden = [500, 100]
dim_output = 1
embed_weights_init = IsotropicGaussian(0.001)
mlp_weights_init = IsotropicGaussian(0.01)
mlp_biases_init = Constant(0.001)
exp_base = 1.5
learning_rate = 0.00001
momentum = 0.99
batch_size = 32
valid_set = 'cuts/test_times_0'
