Mercurial > pylearn

changeset 792:961dc1a7921b

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
added a save and load mecanism and fixed bug to DAA_inputs_groups
author Xavier Glorot <glorotxa@iro.umontreal.ca>
date Fri, 10 Jul 2009 18:14:36 -0400
parents 166a89917669
children 4e70f509ec01
files pylearn/algorithms/sandbox/DAA_inputs_groups.py
diffstat 1 files changed, 90 insertions(+), 13 deletions(-) [+]
line wrap: on
line diff
--- a/pylearn/algorithms/sandbox/DAA_inputs_groups.py	Fri Jul 10 16:56:52 2009 -0400
+++ b/pylearn/algorithms/sandbox/DAA_inputs_groups.py	Fri Jul 10 18:14:36 2009 -0400
@@ -10,6 +10,25 @@
 
 from pylearn.algorithms.logistic_regression import LogRegN
 
+from pylearn.io import filetensor
+import os
+
+# saving loading utils--------------------------------------------
+def save_mat(fname, mat, save_dir=''):
+    assert isinstance(mat, numpy.ndarray)
+    print 'save ndarray to file: ', fname
+    file_handle = open(os.path.join(save_dir, fname), 'w')
+    filetensor.write(file_handle, mat)
+    file_handle.close()
+
+def load_mat(fname, save_dir=''):
+    print 'loading ndarray from file: ', fname
+    file_handle = open(os.path.join(save_dir,fname), 'r')
+    rval = filetensor.read(file_handle)
+    file_handle.close()
+    return rval
+
+
 # Initialize containers:
 class CreateContainer:
     pass
@@ -436,11 +455,11 @@
         auxin_size = auxin_size
         self.n_hid = listify(n_hid,depth)
         self.regularize = regularize
-        self.tie_weights = listify(tie_weights,depth)
-        self.hid_fn = listify(hid_fn,depth)
-        self.rec_fn = listify(rec_fn,depth)
-        self.reconstruction_cost_function = listify(reconstruction_cost_function,depth)
-        self.scale_cost = listify(scale_cost,depth)
+        tie_weights = listify(tie_weights,depth)
+        hid_fn = listify(hid_fn,depth)
+        rec_fn = listify(rec_fn,depth)
+        reconstruction_cost_function = listify(reconstruction_cost_function,depth)
+        scale_cost = listify(scale_cost,depth)
         self.n_out = n_out
         self.target = target if target is not None else T.lvector('target')
         self.debugmethod = debugmethod
@@ -457,11 +476,11 @@
         print '\tauxin_size = ', auxin_size
         print '\tn_hid = ', self.n_hid
         print '\tregularize = ', self.regularize
-        print '\ttie_weights = ', self.tie_weights
-        print '\thid_fn = ', self.hid_fn
-        print '\trec_fn = ', self.rec_fn
-        print '\treconstruction_cost_function = ', self.reconstruction_cost_function
-        print '\tscale_cost = ', self.scale_cost
+        print '\ttie_weights = ', tie_weights
+        print '\thid_fn = ', hid_fn
+        print '\trec_fn = ', rec_fn
+        print '\treconstruction_cost_function = ', reconstruction_cost_function
+        print '\tscale_cost = ', scale_cost
         print '\tn_out = ', self.n_out
         
         # init for model construction
@@ -516,9 +535,9 @@
         for i in range(self.depth):
             
             dict_params = dict(input = inputprec, in_size = in_sizeprec, auxin_size = auxin_size[i],
-                    n_hid = self.n_hid[i], regularize = False, tie_weights = self.tie_weights[i], hid_fn = self.hid_fn[i],
-                    rec_fn = self.rec_fn[i], reconstruction_cost_function = self.reconstruction_cost_function[i],
-                    scale_cost = self.scale_cost[i], interface = False, ignore_missing = self.ignore_missing,
+                    n_hid = self.n_hid[i], regularize = False, tie_weights = tie_weights[i], hid_fn = hid_fn[i],
+                    rec_fn = rec_fn[i], reconstruction_cost_function = reconstruction_cost_function[i],
+                    scale_cost = scale_cost[i], interface = False, ignore_missing = self.ignore_missing,
                     reconstruct_missing = self.reconstruct_missing,corruption_pattern = self.corruption_pattern)
             if auxin_size[i] is None:
                 offset +=1
@@ -671,3 +690,61 @@
                                 low = -1/numpy.sqrt(inst.daaig[-2].n_hid), high = 1/numpy.sqrt(inst.daaig[-2].n_hid))
         inst.daaig[-1].l1 = 0
         inst.daaig[-1].l2 = reg_coef #only l2 norm for regularisation to be consitent with the unsup regularisation
+    
+    def _instance_save(self,inst,save_dir=''):
+        
+        for i in range(self.depth):
+            save_mat('benc%s.ft'%(i) ,inst.daaig[i].benc, save_dir)
+            
+            if self.daaig[i].auxinput is not None:
+                for j in range(len(inst.daaig[i].wauxenc)):
+                    save_mat('wauxenc%s_%s.ft'%(i,j) ,inst.daaig[i].wauxenc[j], save_dir)
+                    save_mat('bauxdec%s_%s.ft'%(i,j) ,inst.daaig[i].bauxdec[j], save_dir)
+            
+            if self.daaig[i].input is not None:
+                save_mat('wenc%s.ft'%(i) ,inst.daaig[i].wenc, save_dir)
+                save_mat('bdec%s.ft'%(i) ,inst.daaig[i].bdec, save_dir)
+            
+            if not self.daaig[i].tie_weights:
+                if self.daaig[i].auxinput is not None:
+                    for j in range(len(inst.daaig[i].wauxdec)):
+                        save_mat('wauxdec%s_%s.ft'%(i,j) ,inst.daaig[i].wauxdec[j], save_dir)
+                
+                if self.daaig[i].input is not None:
+                    save_mat('wdec%s.ft'%(i) ,inst.daaig[i].wdec, save_dir)
+        i=i+1
+        save_mat('wenc%s.ft'%(i) ,inst.daaig[i].w, save_dir)
+        save_mat('benc%s.ft'%(i) ,inst.daaig[i].b, save_dir)
+
+    def _instance_load(self,inst,save_dir='',coef = None, Sup_layer = None):
+        
+        if coef is None:
+            coef = [1]*self.depth
+            
+        for i in range(self.depth):
+            inst.daaig[i].benc = load_mat('benc%s.ft'%(i), save_dir)/coef[i]
+            
+            if self.daaig[i].auxinput is not None:
+                for j in range(len(inst.daaig[i].wauxenc)):
+                    inst.daaig[i].wauxenc[j] = load_mat('wauxenc%s_%s.ft'%(i,j),save_dir)/coef[i]
+                    inst.daaig[i].bauxdec[j] = load_mat('bauxdec%s_%s.ft'%(i,j),save_dir)/coef[i]
+            
+            if self.daaig[i].input is not None:
+                inst.daaig[i].wenc = load_mat('wenc%s.ft'%(i),save_dir)/coef[i]
+                inst.daaig[i].bdec = load_mat('bdec%s.ft'%(i),save_dir)/coef[i]
+            
+            if not self.daaig[i].tie_weights:
+                if self.daaig[i].auxinput is not None:
+                    for j in range(len(inst.daaig[i].wauxdec)):
+                        inst.daaig[i].wauxdec[j] = load_mat('wauxdec%s_%s.ft'%(i,j),save_dir)/coef[i]
+                
+                if self.daaig[i].input is not None:
+                    inst.daaig[i].wdec = load_mat('wdec%s.ft'%(i),save_dir)/coef[i]
+        i=i+1
+        if Sup_layer is None:
+            inst.daaig[i].w = load_mat('wenc%s.ft'%(i),save_dir)
+            inst.daaig[i].b = load_mat('benc%s.ft'%(i),save_dir)
+        else:
+            inst.daaig[i].w = load_mat('wenc%s.ft'%(Sup_layer),save_dir)
+            inst.daaig[i].b = load_mat('benc%s.ft'%(Sup_layer),save_dir)
+