Memory network with bidirectionnal RNN

1 files changed, 56 insertions, 0 deletions

diff --git a/config/memory_network_bidir.py b/config/memory_network_bidir.py
new file mode 100644
index 0000000..dc0824c
--- /dev/null
+++ b/config/memory_network_bidir.py
@@ -0,0 +1,56 @@
+from blocks.initialization import IsotropicGaussian, Constant
+
+from blocks.bricks import Tanh
+
+import data
+from model.memory_network_bidir import Model, Stream
+
+
+dim_embeddings = [
+    ('origin_call', data.origin_call_train_size, 10),
+    ('origin_stand', data.stands_size, 10),
+    ('week_of_year', 52, 10),
+    ('day_of_week', 7, 10),
+    ('qhour_of_day', 24 * 4, 10),
+    ('day_type', 3, 10),
+]
+
+embed_weights_init = IsotropicGaussian(0.001)
+
+
+class RNNConfig(object):
+    __slots__ = ('rec_state_dim', 'dim_embeddings', 'embed_weights_init',
+                 'dim_hidden', 'weights_init', 'biases_init')
+
+prefix_encoder = RNNConfig()
+prefix_encoder.dim_embeddings = dim_embeddings
+prefix_encoder.embed_weights_init = embed_weights_init
+prefix_encoder.rec_state_dim = 100
+prefix_encoder.dim_hidden = [100, 100]
+prefix_encoder.weights_init = IsotropicGaussian(0.01)
+prefix_encoder.biases_init = Constant(0.001)
+
+candidate_encoder = RNNConfig()
+candidate_encoder.dim_embeddings = dim_embeddings
+candidate_encoder.embed_weights_init = embed_weights_init
+candidate_encoder.rec_state_dim = 100
+candidate_encoder.dim_hidden = [100, 100]
+candidate_encoder.weights_init = IsotropicGaussian(0.01)
+candidate_encoder.biases_init = Constant(0.001)
+
+representation_size = 100
+representation_activation = Tanh
+
+normalize_representation = True
+
+
+batch_size = 32
+batch_sort_size = 20
+
+valid_set = 'cuts/test_times_0'
+max_splits = 100
+num_cuts = 1000
+
+train_candidate_size = 100
+valid_candidate_size = 100
+test_candidate_size = 100