# HG changeset patch
# User Pierre-Antoine Manzagol <pierre.antoine.manzagol@gmail.com>
# Date 1265764486 18000
# Node ID 4e3a3d9fef4306bb9ce005551f369955d781cdbf
# Parent  257a39cce72c758d4b8a680072d43d98e9ea20e7
Dataset: added preprocess attribute.

diff -r 257a39cce72c -r 4e3a3d9fef43 pylearn/datasets/dataset.py
--- a/pylearn/datasets/dataset.py	Fri Feb 05 11:53:48 2010 -0500
+++ b/pylearn/datasets/dataset.py	Tue Feb 09 20:14:46 2010 -0500
@@ -108,6 +108,13 @@
 
     img_shape = None # (rows, cols)
 
+    """
+    When inputs 'x' must somehow be preprocessed, processor is a function that
+    will take care of it. 
+    A cleaner (transparent) alternative would be for x to wrap the data intelligently.
+    """
+    preprocess = None
+
 
     """
     TIMESERIES
diff -r 257a39cce72c -r 4e3a3d9fef43 pylearn/datasets/nist_sd.py
--- a/pylearn/datasets/nist_sd.py	Fri Feb 05 11:53:48 2010 -0500
+++ b/pylearn/datasets/nist_sd.py	Tue Feb 09 20:14:46 2010 -0500
@@ -7,6 +7,8 @@
 from pylearn.datasets.config import data_root # config
 from pylearn.datasets.dataset import Dataset
 
+def nist_to_float(x):
+  return x / 255.0
 
 def load(dataset = 'train', attribute = 'data'):
   """Load the filetensor corresponding to the set and attribute.
@@ -35,6 +37,7 @@
   # 
   rval.n_classes = 10
   rval.img_shape = (32,32)
+  rval.preprocess = nist_to_float
 
   # train
   examples = load(dataset = 'train', attribute = 'data')