Mercurial > ift6266

changeset 303:ef28cbb5f464

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Use sigmoids with cross-entropy cost in the ConvAutoencoders.
author Arnaud Bergeron <abergeron@gmail.com>
date Wed, 31 Mar 2010 15:54:47 -0400
parents 1adfafdc3d57
children 1e4bf5a5b46d
files deep/convolutional_dae/scdae.py
diffstat 1 files changed, 4 insertions(+), 2 deletions(-) [+]
line wrap: on
line diff
--- a/deep/convolutional_dae/scdae.py	Tue Mar 30 14:40:54 2010 -0400
+++ b/deep/convolutional_dae/scdae.py	Wed Mar 31 15:54:47 2010 -0400
@@ -14,6 +14,8 @@
                                                    num_filt=num_filt,
                                                    num_in=num_in,
                                                    noisyness=corruption,
+                                                   err=errors.cross_entropy,
+                                                   nlin=nlins.sigmoid,
                                                    dtype=dtype),
                                    MaxPoolLayer(subsampling)])
 
@@ -201,9 +203,9 @@
     pretrain_funcs, trainf, evalf, net = build_funcs(
         img_size = (32, 32),
         batch_size=batch_size, filter_sizes=[(5,5), (3,3)],
-        num_filters=[4, 4], subs=[(2,2), (2,2)], noise=[0.2, 0.2],
+        num_filters=[12, 4], subs=[(2,2), (2,2)], noise=[0.2, 0.2],
         mlp_sizes=[500], out_size=10, dtype=numpy.float32,
-        pretrain_lr=0.01, train_lr=0.1)
+        pretrain_lr=0.001, train_lr=0.1)
     
     t_it = repeat_itf(dset.train, batch_size)
     pretrain_fs, train, valid, test = massage_funcs(