Tensorflowで言語識別をやってみた
Download as PPTX, PDF7 likes12,344 views
12/11、TensorFlowの勉強会で発表した資料
![入力データ
This is a pen.
↓ 文字コード化
[0x54, 0x68, 0x69, 0x73, 0x20, 0x69, 0x73,0x20, 0x61, 0x20, 0x70, 0x65, 0x6e, 0x2e]
・400文字で文章を切って入力](https://image.slidesharecdn.com/whcuper3tu6kghlhkk2f-signature-a35f4e8364a814ed8e355366c11fac0941bc4107f20658d85ec0fb402e725d1a-poli-151211112053/85/Tensorflow-4-320.jpg)
![TensorFlow部分のコード
x = tf.placeholder(tf.float32, [None, n_input_node])
W = tf.Variable(tf.zeros([n_input_node, 3]))
b = tf.Variable(tf.zeros([3]))
y = tf.nn.softmax(tf.matmul(x, W) + b)
y_ = tf.placeholder(tf.float32, [None, 3])
cross_entropy = -tf.reduce_sum(y_*tf.log(y))
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)
init = tf.initialize_all_variables()
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
sess = tf.Session()
sess.run(init)
for i in range(100):
perm = np.random.permutation(n_train)
for i in range(0, n_train, batchsize):
x_batch = np.asarray(x_train[perm[i:i + batchsize]])
y_batch = np.asarray(y_train[perm[i:i + batchsize]])
sess.run(train_step, feed_dict={x: x_batch, y_: y_batch})
print sess.run(accuracy, feed_dict={x: x_test, y_: y_test})](https://image.slidesharecdn.com/whcuper3tu6kghlhkk2f-signature-a35f4e8364a814ed8e355366c11fac0941bc4107f20658d85ec0fb402e725d1a-poli-151211112053/85/Tensorflow-5-320.jpg)
![初心者的TensorFlowの勘所
「y = 5x + 3」を計算してみる
x = tf.placeholder(tf.float32, [None, 1])
y = 5 * x + 3
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
print sess.run(y, feed_dict={x: [[10], [3]]})](https://image.slidesharecdn.com/whcuper3tu6kghlhkk2f-signature-a35f4e8364a814ed8e355366c11fac0941bc4107f20658d85ec0fb402e725d1a-poli-151211112053/85/Tensorflow-8-320.jpg)
![文字だけでなんとかする努力
・入力文字を数値ではなく256次元の1,0のベクトルにしてみた
T
↓ 文字コード化
[0x54]
↓ ベクトル化
[0, 0, 0, …, 0, 0, 1, 0, 0, ….0]](https://image.slidesharecdn.com/whcuper3tu6kghlhkk2f-signature-a35f4e8364a814ed8e355366c11fac0941bc4107f20658d85ec0fb402e725d1a-poli-151211112053/85/Tensorflow-11-320.jpg)
Tensorflowで言語識別をやってみた
- 4. 入力データ This is a pen. ↓ 文字コード化 [0x54, 0x68, 0x69, 0x73, 0x20, 0x69, 0x73,0x20, 0x61, 0x20, 0x70, 0x65, 0x6e, 0x2e] ・400文字で文章を切って入力
- 5. TensorFlow部分のコード x = tf.placeholder(tf.float32, [None, n_input_node]) W = tf.Variable(tf.zeros([n_input_node, 3])) b = tf.Variable(tf.zeros([3])) y = tf.nn.softmax(tf.matmul(x, W) + b) y_ = tf.placeholder(tf.float32, [None, 3]) cross_entropy = -tf.reduce_sum(y_*tf.log(y)) train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy) init = tf.initialize_all_variables() correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float")) sess = tf.Session() sess.run(init) for i in range(100): perm = np.random.permutation(n_train) for i in range(0, n_train, batchsize): x_batch = np.asarray(x_train[perm[i:i + batchsize]]) y_batch = np.asarray(y_train[perm[i:i + batchsize]]) sess.run(train_step, feed_dict={x: x_batch, y_: y_batch}) print sess.run(accuracy, feed_dict={x: x_test, y_: y_test})
- 8. 初心者的TensorFlowの勘所 「y = 5x + 3」を計算してみる x = tf.placeholder(tf.float32, [None, 1]) y = 5 * x + 3 init = tf.initialize_all_variables() sess = tf.Session() sess.run(init) print sess.run(y, feed_dict={x: [[10], [3]]})
- 9. 話を戻して
- 11. 文字だけでなんとかする努力 ・入力文字を数値ではなく256次元の1,0のベクトルにしてみた T ↓ 文字コード化 [0x54] ↓ ベクトル化 [0, 0, 0, …, 0, 0, 1, 0, 0, ….0]