Adding run.sh and rbm.py to numpy/

numpy/rbm.py

@@ -0,0 +1,44 @@
+#!/usr/bin/env python2.5
+from __future__ import absolute_import
+import numpy as N
+import sys
+import time
+
+# c: aa.cc
+
+nin, nout, batchsize, niter = [int(a) for a in sys.argv[1:]]
+lr = 0.01
+
+rng = N.random.RandomState(342)
+
+# declare data
+x = (rng.rand(batchsize*niter, nin)-0.5) * 1.5
+
+# declare model weights
+a = rng.randn(nin) * 0.0
+w = rng.rand(nin, nout)
+b = rng.randn(nout) * 0.0
+
+def sigm(x): return (N.tanh(x)+1)/2
+
+def bern(x): return N.random.binomial(p=x,n=1)
+
+t = time.time()
+for i in xrange(niter):
+    pos_vis = x[i*batchsize:(i+1)*batchsize]
+
+    pos_hid = sigm(N.dot(pos_vis, w)+b)
+
+    neg_vis = sigm(N.dot(bern(pos_hid), w.T)+a)
+
+    neg_hid = sigm(N.dot(bern(neg_vis), w) + b)
+
+    a += lr * N.sum(pos_vis - neg_vis, axis=0)
+    b -= lr * N.sum(pos_hid - neg_hid, axis=0)
+    w -= lr * (N.dot(pos_vis.T, pos_hid) - N.dot(neg_vis.T, neg_hid))
+
+total_time = time.time() - t
+print 'cd1 rbm_bernoulli %i_%i\tnumpy{%i}\t%.2f' %(
+        nin, nout, batchsize, niter*batchsize/total_time)
+
+

numpy/run.sh

@@ -7,3 +7,5 @@ python logreg.py 784 10 1 1000 > ${HOSTNAME}_lr_784_1.bmark
 python logreg.py 784 10 60 100 > ${HOSTNAME}_lr_784_60.bmark
 python logreg.py 32  10 1 1000 > ${HOSTNAME}_lr_32_1.bmark
 python logreg.py 32  10 60 100 > ${HOSTNAME}_lr_32_60.bmark
+python rbm.py 1024 1024 1 100 > ${HOSTNAME}_rbm_1.bmark
+python rbm.py 1024 1024 60 20 > ${HOSTNAME}_rbm_60.bmark