Add the script for training

scripts/mnist-predict-from-model.py

@@ -0,0 +1,93 @@
+
+# coding: utf-8
+
+# In[1]:
+
+import numpy as np
+import tensorflow as tf
+import os
+
+
+# In[2]:
+
+def weight_variable(shape, data=None):
+    initial = data.reshape(shape) if data is not None else tf.truncated_normal(shape, stddev=0.1)
+    return tf.Variable(initial)
+
+def bias_variable(shape, data=None):
+    initial = data.reshape(shape) if data is not None else tf.constant(0.1, shape=shape)
+    return tf.Variable(initial)
+
+def conv2d(x, W):
+    return tf.nn.conv2d(x, W, strides=[1,1,1,1], padding='SAME')
+
+def max_pool_2x2(x):
+    return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')
+
+
+# In[3]:
+
+from tensorflow.examples.tutorials.mnist import input_data
+mnist = input_data.read_data_sets("data/", one_hot=True)
+
+
+# In[4]:
+
+def load(name):
+    return np.fromfile(name, dtype=np.float32)
+
+
+# In[5]:
+
+modeldir = 'model-bnns'
+model_names = ['h1w-5x5x1x32', 'h1b-32', 'h2w-5x5x32x64', 'h2b-64',
+               'h3w-3136x1024', 'h3b-1024', 'h4w-1024x10', 'h4b-10']
+model = [load(os.path.join(modeldir, 'model-' + n)) for n in model_names]
+
+model[0] = model[0].reshape((32,1,5,5)).transpose((2,3,1,0))
+model[2] = model[2].reshape((64,32,5,5)).transpose((2,3,1,0))
+model[4] = model[4].reshape((1024,64,7,7)).transpose((2,3,1,0))
+model[6] = model[6].reshape((10,1024)).transpose()
+
+
+# In[6]:
+
+x = tf.placeholder(tf.float32, [None, 784])
+y_ = tf.placeholder(tf.float32, [None, 10])
+
+x_image = tf.reshape(x, [-1, 28, 28, 1])
+
+W_conv1 = weight_variable([5, 5, 1, 32], model[0])
+b_conv1 = bias_variable([32], model[1])
+h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
+h_pool1 = max_pool_2x2(h_conv1)
+
+W_conv2 = weight_variable([5, 5, 32, 64], model[2])
+b_conv2 = bias_variable([64], model[3])
+h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
+h_pool2 = max_pool_2x2(h_conv2)
+
+W_fc1 = weight_variable([7 * 7 * 64, 1024], model[4])
+b_fc1 = bias_variable([1024], model[5])
+h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])
+h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
+
+W_fc2 = weight_variable([1024, 10], model[6])
+b_fc2 = bias_variable([10], model[7])
+
+y_conv = tf.nn.softmax(tf.matmul(h_fc1, W_fc2) + b_fc2)
+
+
+# In[7]:
+
+sample = mnist.test.images[0:1].copy()
+actual = mnist.test.labels[0].argmax()
+
+init = tf.initialize_all_variables()
+with tf.Session() as ses:
+    ses.run(init)
+    predict = tf.argmax(y_conv, 1)
+    recognized = ses.run(predict, feed_dict = {x: sample})
+
+'recognized %d, actual %d' % (recognized, actual)
+