i forget

Author: lazyprogrammer

.gitignore

@@ -1,3 +1,4 @@
+*.DS_Store
 *.pyc
 large_files
 large_files/*

ann_class2/rmsprop.py

@@ -82,7 +82,7 @@ def main():
     cache_W1 = 0
     cache_b1 = 0
     decay_rate = 0.999
-    eps = 0.000001
+    eps = 0.0000000001
     for i in xrange(max_iter):
         for j in xrange(n_batches):
             Xbatch = Xtrain[j*batch_sz:(j*batch_sz + batch_sz),]

ann_class2/tensorflow2.py

@@ -26,7 +26,7 @@ def main():
     print_period = 10
 
     lr = 0.00004
-    mu = 0.9
+    reg = 0.01
 
     Xtrain = X[:-1000,]
     Ytrain = Y[:-1000]
@@ -94,7 +94,7 @@ def main():
                     test_cost = session.run(cost, feed_dict={X: Xtest, T: Ytest_ind})
                     prediction = session.run(predict_op, feed_dict={X: Xtest})
                     err = error_rate(prediction, Ytest)
-                    print "Cost / err at iteration i=%d, j=%d: %.6f / %.3f" % (i, j, test_cost, err)
+                    print "Cost / err at iteration i=%d, j=%d: %.3f / %.3f" % (i, j, test_cost, err)
                     LL.append(test_cost)
 
     plt.plot(LL)

ann_class2/util.py

@@ -206,7 +206,7 @@ def benchmark_pca():
     # D = 300 -> error = 0.07
     lr = 0.0001
     reg = 0.01
-    for i in xrange(800):
+    for i in xrange(200):
         p_y = forward(Xtrain, W, b)
         # print "p_y:", p_y
         ll = cost(p_y, Ytrain_ind)
@@ -235,6 +235,6 @@ def benchmark_pca():
 
 
 if __name__ == '__main__':
-    # benchmark_pca()
-    benchmark_full()
+    benchmark_pca()
+    # benchmark_full()
 

logistic_regression_class/logistic_donut.py

@@ -53,13 +53,14 @@ def sigmoid(z):
 
 # calculate the cross-entropy error
 def cross_entropy(T, Y):
-    E = 0
-    for i in xrange(N):
-        if T[i] == 1:
-            E -= np.log(Y[i])
-        else:
-            E -= np.log(1 - Y[i])
-    return E
+    # E = 0
+    # for i in xrange(N):
+    #     if T[i] == 1:
+    #         E -= np.log(Y[i])
+    #     else:
+    #         E -= np.log(1 - Y[i])
+    # return E
+    return (T*np.log(Y) + (1-T)*np.log(1-Y)).sum()
 
 
 # let's do gradient descent 100 times