# HG changeset patch
# User fsavard
# Date 1267559238 18000
# Node ID 3632e6258642499640868362bfe9bd294429b5c2
# Parent  70a9df1cd20e7232d524c9e921973fc4f98ff58c
Ajouts mineurs à stacked_dae, juste printé l'heure je crois.

diff -r 70a9df1cd20e -r 3632e6258642 deep/stacked_dae/mnist_sda.py
--- a/deep/stacked_dae/mnist_sda.py	Tue Mar 02 09:52:27 2010 -0500
+++ b/deep/stacked_dae/mnist_sda.py	Tue Mar 02 14:47:18 2010 -0500
@@ -1,6 +1,7 @@
 #!/usr/bin/python
 # coding: utf-8
 
+# TODO: This probably doesn't work anymore, adapt to new code in sgd_opt
 # Parameterize call to sgd_optimization for MNIST
 
 import numpy 
diff -r 70a9df1cd20e -r 3632e6258642 deep/stacked_dae/nist_sda.py
--- a/deep/stacked_dae/nist_sda.py	Tue Mar 02 09:52:27 2010 -0500
+++ b/deep/stacked_dae/nist_sda.py	Tue Mar 02 14:47:18 2010 -0500
@@ -25,18 +25,13 @@
 
 from sgd_optimization import SdaSgdOptimizer
 
-SERIES_AVAILABLE = False
-try:
-    from scalar_series import *
-    SERIES_AVAILABLE = True
-except ImportError:
-    print "Could not import Series"
+from ift6266.utils.scalar_series import *
 
 TEST_CONFIG = False
 
 NIST_ALL_LOCATION = '/data/lisa/data/nist/by_class/all'
 
-JOBDB = 'postgres://ift6266h10@gershwin/ift6266h10_db/fsavard_sda2'
+JOBDB = 'postgres://ift6266h10@gershwin/ift6266h10_sandbox_db/fsavard_sda2'
 
 REDUCE_TRAIN_TO = None
 MAX_FINETUNING_EPOCHS = 1000
@@ -58,15 +53,15 @@
         'num_hidden_layers':[2,3]}
 
 # Just useful for tests... minimal number of epochs
-DEFAULT_HP_NIST = DD({'finetuning_lr':0.01,
-                       'pretraining_lr':0.01,
-                       'pretraining_epochs_per_layer':1,
-                       'max_finetuning_epochs':1,
-                       'hidden_layers_sizes':1000,
+DEFAULT_HP_NIST = DD({'finetuning_lr':0.1,
+                       'pretraining_lr':0.1,
+                       'pretraining_epochs_per_layer':20,
+                       'max_finetuning_epochs':2,
+                       'hidden_layers_sizes':300,
                        'corruption_levels':0.2,
                        'minibatch_size':20,
-                       'reduce_train_to':1000,
-                       'num_hidden_layers':1})
+                       #'reduce_train_to':300,
+                       'num_hidden_layers':2})
 
 def jobman_entrypoint(state, channel):
     pylearn.version.record_versions(state,[theano,ift6266,pylearn])
@@ -75,12 +70,10 @@
     workingdir = os.getcwd()
 
     print "Will load NIST"
-    sys.stdout.flush()
 
     nist = NIST(20)
 
     print "NIST loaded"
-    sys.stdout.flush()
 
     rtt = None
     if state.has_key('reduce_train_to'):
@@ -89,7 +82,7 @@
         rtt = REDUCE_TRAIN_TO
 
     if rtt:
-        print "Reducing training set to ", rtt, " examples"
+        print "Reducing training set to "+str( rtt)+ " examples"
         nist.reduce_train_set(rtt)
 
     train,valid,test = nist.get_tvt()
@@ -107,8 +100,9 @@
     # b,b',W for each hidden layer + b,W of last layer (logreg)
     numparams = nhl * 3 + 2
     series_mux = None
-    if SERIES_AVAILABLE:
-        series_mux = create_series(workingdir, numparams)
+    series_mux = create_series(workingdir, numparams)
+
+    print "Creating optimizer with state, ", state
 
     optimizer = SdaSgdOptimizer(dataset=dataset, hyperparameters=state, \
                                     n_ins=n_ins, n_outs=n_outs,\
@@ -275,7 +269,9 @@
         jobman_insert_nist()
 
     elif len(args) > 0 and args[0] == 'test_jobman_entrypoint':
-        chanmock = DD({'COMPLETE':0})
+        def f():
+            pass
+        chanmock = DD({'COMPLETE':0,'save':f})
         jobman_entrypoint(DEFAULT_HP_NIST, chanmock)
 
     elif len(args) > 0 and args[0] == 'estimate':
diff -r 70a9df1cd20e -r 3632e6258642 deep/stacked_dae/sgd_optimization.py
--- a/deep/stacked_dae/sgd_optimization.py	Tue Mar 02 09:52:27 2010 -0500
+++ b/deep/stacked_dae/sgd_optimization.py	Tue Mar 02 14:47:18 2010 -0500
@@ -6,6 +6,7 @@
 import numpy 
 import theano
 import time
+import datetime
 import theano.tensor as T
 import sys
 
@@ -85,7 +86,7 @@
         self.finetune()
 
     def pretrain(self):
-        print "STARTING PRETRAINING"
+        print "STARTING PRETRAINING, time = ", datetime.datetime.now()
         sys.stdout.flush()
 
         start_time = time.clock()  
@@ -101,6 +102,8 @@
                         
                 print 'Pre-training layer %i, epoch %d, cost '%(i,epoch),c
                 sys.stdout.flush()
+
+                self.series_mux.append("params", self.classifier.all_params)
      
         end_time = time.clock()
 
@@ -110,7 +113,7 @@
         sys.stdout.flush()
 
     def finetune(self):
-        print "STARTING FINETUNING"
+        print "STARTING FINETUNING, time = ", datetime.datetime.now()
 
         index   = T.lscalar()    # index to a [mini]batch 
         minibatch_size = self.hp.minibatch_size
diff -r 70a9df1cd20e -r 3632e6258642 deep/stacked_dae/stacked_dae.py
--- a/deep/stacked_dae/stacked_dae.py	Tue Mar 02 09:52:27 2010 -0500
+++ b/deep/stacked_dae/stacked_dae.py	Tue Mar 02 14:47:18 2010 -0500
@@ -138,8 +138,6 @@
     self.params = [ self.W, self.b, self.b_prime ]
 
 
-
-
 class SdA(object):
     def __init__(self, train_set_x, train_set_y, batch_size, n_ins, 
                  hidden_layers_sizes, n_outs, 
@@ -147,6 +145,7 @@
         # Just to make sure those are not modified somewhere else afterwards
         hidden_layers_sizes = copy.deepcopy(hidden_layers_sizes)
         corruption_levels = copy.deepcopy(corruption_levels)
+
         update_locals(self, locals())      
  
         self.layers             = []
diff -r 70a9df1cd20e -r 3632e6258642 deep/stacked_dae/utils.py
--- a/deep/stacked_dae/utils.py	Tue Mar 02 09:52:27 2010 -0500
+++ b/deep/stacked_dae/utils.py	Tue Mar 02 14:47:18 2010 -0500
@@ -1,4 +1,7 @@
 #!/usr/bin/python
+# coding: utf-8
+
+from __future__ import with_statement
 
 from jobman import DD