taxi

commit 6cf80aff3e4dc57d13a33c2946bc0ae57cfae6b8
parent 5096e0cdae167122d07b09cd207a04f28ea5c3f5
Author: Alex Auvolat <alex.auvolat@ens.fr>
Date:   Thu,  2 Jul 2015 14:32:23 -0400

More determinism

Diffstat:
M
data/transformers.py
 | 
1
+
M
model/mlp.py
 | 
5
+++++
M
train.py
 | 
18
+++++++++++-------

3 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/data/transformers.py b/data/transformers.py
@@ -10,6 +10,7 @@ from fuel.transformers import Batch, Mapping, SortMapping, Transformer, Unpack
 
 import data
 
+fuel.config.default_seed = 123
 
 def at_least_k(k, v, pad_at_begin, is_longitude):
     if len(v) == 0:
diff --git a/model/mlp.py b/model/mlp.py
@@ -1,5 +1,8 @@
 from theano import tensor
 
+import fuel
+import blocks
+
 from fuel.transformers import Batch, MultiProcessing
 from fuel.streams import DataStream
 from fuel.schemes import ConstantScheme, ShuffledExampleScheme
@@ -11,6 +14,8 @@ from data.hdf5 import TaxiDataset, TaxiStream
 from data.cut import TaxiTimeCutScheme
 from model import ContextEmbedder
 
+blocks.config.default_seed = 123
+fuel.config.default_seed = 123
 
 class FFMLP(Initializable):
     def __init__(self, config, output_layer=None, **kwargs):
diff --git a/train.py b/train.py
@@ -87,6 +87,9 @@ if __name__ == "__main__":
     else:
         step_rule = AdaDelta()
 
+    logger.info("Fuel seed: %d" % fuel.config.default_seed)
+    logger.info("Blocks seed: %d" % blocks.config.default_seed)
+
     params = cg.parameters
     algorithm = GradientDescent(
         cost=cost,
@@ -107,13 +110,14 @@ if __name__ == "__main__":
                                      prefix='valid',
                                      every_n_batches=1000),
                 Printing(every_n_batches=1000),
-
-                # SaveLoadParams(dump_path, cg,
-                #                before_training=True,        # before training -> load params
-                #                every_n_batches=1000,        # every N batches -> save params
-                #                after_epoch=True,            # after epoch -> save params
-                #                after_training=True,         # after training -> save params
-                #                ),
+                # FinishAfter(every_n_batches=10),
+
+                SaveLoadParams(dump_path, cg,
+                               before_training=True,        # before training -> load params
+                               every_n_batches=1000,        # every N batches -> save params
+                               after_epoch=True,            # after epoch -> save params
+                               after_training=True,         # after training -> save params
+                               ),
 
                 RunOnTest(model_name,
                           model,