Character-level Convolutional Networks for Text Classification论文仿真实现

Home Page: http://blog.csdn.net/liuchonge/article/details/70947995

Python 100.00%

cnn tensorflow char-cnn-rnn text-classification

char-cnn-text-classification-tensorflow's Introduction

char-cnn-text-classification-tensorflow

主要用于记录“Character-level Convolutional Networks for Text Classification”论文的模型架构和仿真实现方法。这是一篇2016年4月份刚发的文章，在此之前，原作者还发表过一篇“Text Understanding from Scratch”的论文，两篇论文基本上是一样的。可以参见自己的博客http://blog.csdn.net/liuchonge/article/details/70947995

char-cnn-text-classification-tensorflow's People

Contributors

Stargazers

Watchers

char-cnn-text-classification-tensorflow's Issues

关于测试的文件

能请博主上传一下测试的相关代码吗？
另外还有一个问题是：论文中一共有9层，代码里只看到了卷积加上全链接7层？
deep learning 的新手，还请指导！

Error: Argument must be a dense tensor:

ValueError: Tried to convert 'indices' to a tensor and failed. Error: Argument must be a dense tensor: range(0, 69) - got shape [69], but wanted [].
博主有没有遇到这类问题？

Sudden increase of loss (from 0.1 to thousands) after ~4000 steps

I got a sudden drop of accuracy from ~90% to below 20% after about 4000 training steps? An answer on stack overflow says that's a problem of Adam, and suggests a higher epsilon to fix?

Do you have any suggestion based on your training experience? Thanks!

My Stats:

...
2018-03-21T13:56:22.703288: step 3866, loss 0.43315, acc 0.90625
2018-03-21T13:56:34.078052: step 3867, loss 0.479325, acc 0.851562
2018-03-21T13:56:49.068364: step 3868, loss 0.256937, acc 0.9375
2018-03-21T13:57:06.444321: step 3869, loss 0.273636, acc 0.90625
2018-03-21T13:57:15.877391: step 3870, loss 0.232426, acc 0.929688
2018-03-21T13:57:23.503175: step 3871, loss 3.60959, acc 0.5625
2018-03-21T13:57:31.559549: step 3872, loss 8.54077, acc 0.46875
2018-03-21T13:57:41.647740: step 3873, loss 5.25972, acc 0.429688
2018-03-21T13:57:51.344488: step 3874, loss 322.297, acc 0.234375
2018-03-21T13:58:01.384422: step 3875, loss 21.8541, acc 0.453125
2018-03-21T13:58:10.741925: step 3876, loss 131.247, acc 0.242188
2018-03-21T13:58:19.820908: step 3877, loss 4977.22, acc 0.289062
2018-03-21T13:58:29.067122: step 3878, loss 155.665, acc 0.289062
2018-03-21T13:58:36.979536: step 3879, loss 1825.98, acc 0.226562
2018-03-21T13:58:44.365748: step 3880, loss 1705.01, acc 0.304688
2018-03-21T13:58:53.205122: step 3881, loss 1100.61, acc 0.25
2018-03-21T13:59:01.279594: step 3882, loss 12413.7, acc 0.234375
2018-03-21T13:59:10.017025: step 3883, loss 23248.9, acc 0.273438
...

我想restore预测单个句子，应该怎么弄呢？

报错：FailedPreconditionError (see above for traceback): Attempting to use uninitialized value conv_layer-1/w
[[Node: conv_layer-1/w/read = IdentityT=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:GPU:0"]]
[[Node: output_layer/predictions/_5 = _Recvclient_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_157_output_layer/predictions", tensor_type=DT_INT64, _device="/job:localhost/replica:0/task:0/device:CPU:0"]]

predition代码:

coding=utf-8

import tensorflow as tf
from NN_Real_examples.char_CNN_text_classification_tf.data_helper import Dataset
import time
import os
from tensorflow.python import debug as tf_debug
from NN_Real_examples.char_CNN_text_classification_tf.charCNN import CharCNN
import datetime
from NN_Real_examples.char_CNN_text_classification_tf.config import config
import numpy as np

Load data

print("正在载入数据...")
dev_data = Dataset(config.dev_data_source)

print ("得到120000维的doc_train，label_train")
print ("得到9600维的doc_dev, label_train")

with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=False)
sess = tf.Session(config=session_conf)
# sess = tf_debug.LocalCLIDebugWrapperSession(sess)
with sess.as_default():
cnn = CharCNN(
l0=config.l0,
num_classes=config.nums_classes,
conv_layers=config.model.conv_layers,
fc_layers=config.model.fc_layers,
l2_reg_lambda=0)

    saver = tf.train.Saver()
    module_file = tf.train.latest_checkpoint(
        'D:/workspace/myPython/NN_Real_examples/char_CNN_text_classification_tf/runs/1530092351/checkpoints/model-3000.data-00000-of-00001')  # ckpt路径抽调出来
    sess = tf.Session()
    if module_file is not None:  # 添加一个判断语句，判断ckpt的路径文件
        saver.restore(sess, save_path=module_file)  # 读取保存的模型

    def dev_step(x_batch, y_batch):
        feed_dict = {
          cnn.input_x: x_batch,
          cnn.input_y: y_batch,
          cnn.dropout_keep_prob: 1.0
        }
        predictions = sess.run(
            [cnn.predictions],
            feed_dict)
        return predictions

    try:
        text = "Wall St. Bears Claw Back Into the Black (Reuters).Reuters - Short-sellers, Wall Street's dwindling\band of ultra-cynics, are seeing green again."
        embedding_w, embedding_dic = dev_data.onehot_dic_build()
        doc_vec = dev_data.doc_process(text, embedding_dic)
        doc_image = []
        label_image = []
        label_class = np.zeros(4, dtype='float32')
        label_class[0] = 1
        doc_image.append(doc_vec)
        label_image.append(label_class)
        batch_x = np.array(doc_image, dtype='int64')
        batch_y = np.array(label_image, dtype='float32')
        predictions = dev_step(batch_x, batch_y)
        label_pred = predictions[0].tolist()  # 转换为list
        print(label_pred)
    except EOFError:
        print("Closing session.")

lc222 / char-cnn-text-classification-tensorflow Goto Github PK

char-cnn-text-classification-tensorflow's Introduction

char-cnn-text-classification-tensorflow

char-cnn-text-classification-tensorflow's People

Contributors

Stargazers

Watchers

Forkers

char-cnn-text-classification-tensorflow's Issues

关于测试的文件

Error: Argument must be a dense tensor:

Sudden increase of loss (from 0.1 to thousands) after ~4000 steps

我想restore预测单个句子，应该怎么弄呢？

coding=utf-8

Load data

您好有仿真的代码吗

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent