memory_network_bidir_momentum.py (1581B)
1 from blocks.initialization import IsotropicGaussian, Constant 2 from blocks.algorithms import Momentum 3 4 from blocks.bricks import Tanh 5 6 import data 7 from model.memory_network_bidir import Model, Stream 8 9 10 dim_embeddings = [ 11 ('origin_call', data.origin_call_train_size, 10), 12 ('origin_stand', data.stands_size, 10), 13 ('week_of_year', 52, 10), 14 ('day_of_week', 7, 10), 15 ('qhour_of_day', 24 * 4, 10), 16 ('day_type', 3, 10), 17 ] 18 19 embed_weights_init = IsotropicGaussian(0.001) 20 21 22 class RNNConfig(object): 23 __slots__ = ('rec_state_dim', 'dim_embeddings', 'embed_weights_init', 24 'dim_hidden', 'weights_init', 'biases_init') 25 26 prefix_encoder = RNNConfig() 27 prefix_encoder.dim_embeddings = dim_embeddings 28 prefix_encoder.embed_weights_init = embed_weights_init 29 prefix_encoder.rec_state_dim = 100 30 prefix_encoder.dim_hidden = [100, 100] 31 prefix_encoder.weights_init = IsotropicGaussian(0.01) 32 prefix_encoder.biases_init = Constant(0.001) 33 34 candidate_encoder = RNNConfig() 35 candidate_encoder.dim_embeddings = dim_embeddings 36 candidate_encoder.embed_weights_init = embed_weights_init 37 candidate_encoder.rec_state_dim = 100 38 candidate_encoder.dim_hidden = [100, 100] 39 candidate_encoder.weights_init = IsotropicGaussian(0.01) 40 candidate_encoder.biases_init = Constant(0.001) 41 42 representation_size = 100 43 representation_activation = Tanh 44 45 normalize_representation = True 46 47 48 batch_size = 64 49 batch_sort_size = 20 50 51 max_splits = 100 52 num_cuts = 1000 53 54 train_candidate_size = 100 55 valid_candidate_size = 100 56 test_candidate_size = 100 57 58 step_rule = Momentum(learning_rate=0.01, momentum=0.9)