TensorFlow分batch加載並生成數據

import tensorflow as tf
import pandas as pd
import time

sess = tf.InteractiveSession()


def gen():
    csv_data = pd.read_csv('../dataset/train.csv')
    length = csv_data.shape[0]
    for i in range(length):
        [id, tokens, label] = csv_data.iloc[i]
        yield (id, tokens, label)


def create_dataset():
    data = tf.data.Dataset.from_generator(gen, (tf.int32, tf.string, tf.string))
    data = data.batch(32)
    data = data.prefetch(32)
    data = data.make_one_shot_iterator()
    get_id, get_tokens, get_label = data.get_next()
    while True:
        try:
            id, tokens, label = sess.run([get_id, get_tokens, get_label])
            yield id, tokens, label
        except:
            break


data = create_dataset()
for i, (id, tokens, label) in enumerate(data):
    # print(id, tokens, label)
    print(id.shape[0])
    print('----------------')