Pythonによるスクレイピング&機械学習[6-4] #TensorFlow

MLPでテキスト分類してみよう

6章で、テキスト分類をするコードがkerasで書かれていたので、tensorflowで書き直しました。(p290)

コード

import tensorflow as tf
import json
from sklearn.model_selection import train_test_split
from keras.utils import np_utils
from sklearn.utils import shuffle

nb_classes = 9

batch_size = 64
epochs = 20

def weight_variable(name, shape):
    W_init = tf.truncated_normal(shape, stddev=0.01)
    W = tf.Variable(W_init, name="W_"+name)
    return W

def bias_variable(name, size):
    b_init = tf.constant(0.1, shape=[size])
    b = tf.Variable(b_init, name="b_"+name)
    return b

# data = json.load(open("./newstext/data-mini.json"))
data = json.load(open("./newstext/data.json"))
X = data["X"]
Y = data["Y"]
max_words = len(X[0])

X_train, X_test, Y_train, Y_test = train_test_split(X, Y)
Y_train = np_utils.to_categorical(Y_train, nb_classes)
Y_test = np_utils.to_categorical(Y_test, nb_classes)
print(len(X_train), len(Y_train))

x = tf.placeholder(tf.float32, shape=[None, max_words])
y_ = tf.placeholder(tf.float32, shape=[None, nb_classes])

W = weight_variable("num1",[max_words, 512])
b = bias_variable("num1", 512)
h = tf.nn.relu(tf.matmul(x, W) + b)

keep_prob = tf.placeholder(tf.float32)
h_drop = tf.nn.dropout(h, keep_prob)

W2 = weight_variable("num2", [512, nb_classes])
b2 = bias_variable("num2", nb_classes)
y = tf.nn.softmax(tf.matmul(h_drop, W2) + b2)

cross_entoropy = -tf.reduce_sum(y_ * tf.log(tf.clip_by_value(y, 1e-10, 1.0)))

optimizer = tf.train.AdamOptimizer(1e-4)
train_step = optimizer.minimize(cross_entoropy)

predict_step = tf.equal(tf.argmax(y_, 1), tf.argmax(y, 1))
accuracy_step = tf.reduce_mean(tf.cast(predict_step, tf.float32))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    N_train = len(X_train)
    n_batches = N_train // batch_size
    for epoch in range(epochs):
        X_, Y_ = shuffle(X_train, Y_train)
        for i in range(n_batches):
            start = i * batch_size
            end = start + batch_size

            sess.run(train_step, feed_dict={
                x: X_[start:end],
                y_:Y_[start:end],
                keep_prob:0.5
            })

        loss = sess.run(cross_entoropy, feed_dict={
            x: X_test,
            y_: Y_test,
            keep_prob:1
        })
        print(loss)

        accuracy = sess.run(accuracy_step, feed_dict={
            x: X_test,
            y_: Y_test,
            keep_prob:1
        })
        print(accuracy)

    acc = sess.run(accuracy_step, feed_dict={
        x: X_test,
        y_:Y_test,
        keep_prob:1.0
    })
    print(acc)

結果

参照

Pythonによるスクレイピング＆機械学習　開発テクニック BeautifulSoup、scikit-learn、TensorFlowを使ってみよう(著者:クジラ飛行机、出版社:ソシム株式会社)
http://www.socym.co.jp/book/1079