Mnist updates

examples/mnist_full.jl

@@ -1,4 +1,4 @@
-# using TensorFlow
+using TensorFlow
 using Distributions
 include("mnist_loader.jl")
 
@@ -24,38 +24,41 @@ function max_pool_2x2(x)
     nn.max_pool(x, [1, 2, 2, 1], [1, 2, 2, 1], "SAME")
 end
 
-x = placeholder(Float32)
-y_ = placeholder(Float32)
+@tf begin
 
-W_conv1 = weight_variable([5, 5, 1, 32])
-b_conv1 = bias_variable([32])
+    x = placeholder(Float32)
+    y_ = placeholder(Float32)
 
-x_image = reshape(x, [-1, 28, 28, 1])
+    W_conv1 = weight_variable([5, 5, 1, 32])
+    b_conv1 = bias_variable([32])
 
-h_conv1 = nn.relu(conv2d(x_image, W_conv1) + b_conv1)
-h_pool1 = max_pool_2x2(h_conv1)
+    x_image = reshape(x, [-1, 28, 28, 1])
 
-W_conv2 = weight_variable([5, 5, 32, 64])
-b_conv2 = bias_variable([64])
+    h_conv1 = nn.relu(conv2d(x_image, W_conv1) + b_conv1)
+    h_pool1 = max_pool_2x2(h_conv1)
 
-h_conv2 = nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
-h_pool2 = max_pool_2x2(h_conv2)
+    W_conv2 = weight_variable([5, 5, 32, 64])
+    b_conv2 = bias_variable([64])
 
-W_fc1 = weight_variable([7*7*64, 1024])
-b_fc1 = bias_variable([1024])
+    h_conv2 = nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
+    h_pool2 = max_pool_2x2(h_conv2)
 
-h_pool2_flat = reshape(h_pool2, [-1, 7*7*64])
-h_fc1 = nn.relu(h_pool2_flat * W_fc1 + b_fc1)
+    W_fc1 = weight_variable([7*7*64, 1024])
+    b_fc1 = bias_variable([1024])
 
-keep_prob = placeholder(Float32)
-h_fc1_drop = nn.dropout(h_fc1, keep_prob)
+    h_pool2_flat = reshape(h_pool2, [-1, 7*7*64])
+    h_fc1 = nn.relu(h_pool2_flat * W_fc1 + b_fc1)
 
-W_fc2 = weight_variable([1024, 10])
-b_fc2 = bias_variable([10])
+    keep_prob = placeholder(Float32)
+    h_fc1_drop = nn.dropout(h_fc1, keep_prob)
 
-y_conv = nn.softmax(h_fc1_drop * W_fc2 + b_fc2)
+    W_fc2 = weight_variable([1024, 10])
+    b_fc2 = bias_variable([10])
 
-cross_entropy = reduce_mean(-reduce_sum(y_.*log(y_conv), axis=[2]))
+    global y_conv = nn.softmax(h_fc1_drop * W_fc2 + b_fc2)
+
+    global cross_entropy = reduce_mean(-reduce_sum(y_.*log(y_conv), axis=[2]))
+end
 
 train_step = train.minimize(train.AdamOptimizer(1e-4), cross_entropy)
 
@@ -65,7 +68,7 @@ accuracy = reduce_mean(cast(correct_prediction, Float32))
 
 run(session, global_variables_initializer())
 
-for i in 1:1000
+for i in 1:200
     batch = next_batch(loader, 50)
     if i%100 == 1
         train_accuracy = run(session, accuracy, Dict(x=>batch[1], y_=>batch[2], keep_prob=>1.0))
@@ -77,3 +80,5 @@ end
 testx, testy = load_test_set()
 test_accuracy = run(session, accuracy, Dict(x=>testx, y_=>testy, keep_prob=>1.0))
 info("test accuracy $test_accuracy")
+
+visualize()