<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[cjopengler的专栏]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/cjopengler</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; cjopengler]]></copyright><item><title><![CDATA[allennlp jsonnet  mac 安装失败解决方案]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/100555128</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/100555128</guid><author>cjopengler</author><pubDate>Thu, 05 Sep 2019 11:43:43 +0800</pubDate><description><![CDATA[安装命令: pip install jsonnet
报错如下:
lang: warning: libstdc++ is deprecated; move to libc++ with a minimum deployment target of OS X 10.9 [-Wdeprecated] ld: library not found for -lstdc++ clang: error: lin...]]></description><category></category></item><item><title><![CDATA[NLTK 入门]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/80662873</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/80662873</guid><author>cjopengler</author><pubDate>Tue, 12 Jun 2018 11:19:03 +0800</pubDate><description><![CDATA[NLTK 入门



from matplotlib import pyplot as plt
from nltk import book

*** Introductory Examples for the NLTK Book ***
Loading text1, ..., text9 and sent1, ..., sent9
Type the name of the text or sent...]]></description><category></category></item><item><title><![CDATA[pandas 时间index创建与使用]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/80649285</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/80649285</guid><author>cjopengler</author><pubDate>Mon, 11 Jun 2018 11:15:35 +0800</pubDate><description><![CDATA[# 时间index

在处理某些数据的时候，比如股票数据，当以时间进行索引会容易处理很多。








import pandas as pd









stock = {'open': [1.0, 1.1, 1.2], 'close': [1.5, 1.9, 0.7]}
index = ['2018-03-04', '2018-03-05', '2018-03-06']

# 这样三...]]></description><category></category></item><item><title><![CDATA[马尔科夫链 吸收概率方程组推导]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78811396</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78811396</guid><author>cjopengler</author><pubDate>Fri, 15 Dec 2017 11:54:56 +0800</pubDate><description><![CDATA[利用了全概率公式
马尔科夫链的相关性质]]></description><category></category></item><item><title><![CDATA[python 正则表达式 (持续更新)]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78412071</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78412071</guid><author>cjopengler</author><pubDate>Wed, 01 Nov 2017 11:18:32 +0800</pubDate><description><![CDATA[正则表达式import re中文source = u"s2f程序员杂志一2d3程序员杂志二2d3程序员杂志三2d3程序员杂志四2d3"temp = source
xx = u"([\u4e00-\u9fa5]+)"
pattern = re.compile(xx)
results = pattern.findall(temp)print(len(results))
for result in res]]></description><category></category></item><item><title><![CDATA[第十三课 wide&deep模型]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78284643</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78284643</guid><author>cjopengler</author><pubDate>Thu, 19 Oct 2017 14:39:44 +0800</pubDate><description><![CDATA[这篇paper很简单，网上也有很多人翻译过来. 使用tensorflow自带的库，其实很简单。有些难点的地方是，关于特征工程部分的理解。请参考 第九课 tensorflow 特征工程: feature_column下面是具体的实践demo：# coding:utf-8
"""
wide and deep
"""from framework.data_input import IDataInput
f]]></description><category></category></item><item><title><![CDATA[第十二课 tensorflow 使用RNN实现古诗自动生成]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78273923</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78273923</guid><author>cjopengler</author><pubDate>Wed, 18 Oct 2017 15:57:18 +0800</pubDate><description><![CDATA[上一课中说到RNN的实现原理。这一章，一个古诗生成的demo.输入# coding:utf-8
"""
数据输入
"""import logging
import collections
import json
import numpy as np
class PoemInput(object):    def __init__(self, poem_file_path, batch_size):]]></description><category></category></item><item><title><![CDATA[tensorflow 调试技巧汇总]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78273684</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78273684</guid><author>cjopengler</author><pubDate>Wed, 18 Oct 2017 15:43:47 +0800</pubDate><description><![CDATA[learning rate当发现loss下降，然后又开始上升。很可能是learning rate 到后面太大导致的缘故。 
所以应该使用带有衰减效果的learning rate.session.run(tf.assign(learning_rate, 0.002 * (0.97 ** (step // decay_steps))))]]></description><category></category></item><item><title><![CDATA[python中文输出]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78247208</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78247208</guid><author>cjopengler</author><pubDate>Mon, 16 Oct 2017 10:38:30 +0800</pubDate><description><![CDATA[# coding:utfimport json
def test_dict_output():
    a_dict = {'a': u'你好',
              'b': u'中国'}
    a_dict_str = json.dumps(a_dict, ensure_ascii=False).encode('utf-8')
    print(a_dict_str)
def tes]]></description><category></category></item><item><title><![CDATA[第十一课 tensorflow RNN原理及解析]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78227004</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78227004</guid><author>cjopengler</author><pubDate>Fri, 13 Oct 2017 15:56:42 +0800</pubDate><description><![CDATA[RNN原理解析RNN的一层结构图如下:
W: state 权重
U: 输入权重
V: 输出权重
xt: t个step的输入
ht: rnn cell 隐层, 也有的叫st (状态层)
ot: 最终的输出
yt: 经过soft max之后的分类结果
数学关系: 
 
关于维度: 
* ht维度: 是隐层的数量，也是自定义的, shape=[hn✖️1] 
* W维度: shape=[hn, hn],才]]></description><category></category></item><item><title><![CDATA[tensorflow 特征工程中的combiner]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78206273</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78206273</guid><author>cjopengler</author><pubDate>Wed, 11 Oct 2017 17:32:55 +0800</pubDate><description><![CDATA[combinercombiner的行为tf.contrib.layers.sparse_column_with_hash_bucket 的参数中有一个combiner. 他的含义是说，当产生稠密向量的时候，使用combiner与权重重重新计算权重，也就是相当于对权重的正则化。举个例子来说，feature A中，包含两个稀疏向量，分别是[0, 1, 0]和[0,0,1] 他们分别对应一个权重0.4和0]]></description><category></category></item><item><title><![CDATA[第十课 tensorflow 实现word2vec]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78189922</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78189922</guid><author>cjopengler</author><pubDate>Tue, 10 Oct 2017 10:37:12 +0800</pubDate><description><![CDATA[精简版本，去掉了各种tsne以及vector可视化的麻烦，直接看到word2vec的训练过程对文本进行预处理也就是将每一个单词转换成一个数字序号，同时将低频词过滤掉，统一使用 UKN来代替.import collections
import tensorflow as tf
def pre_process(input_file_path, vocabulary_size, output_file_p]]></description><category></category></item><item><title><![CDATA[tensorflow 常见错误及解析]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78161759</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78161759</guid><author>cjopengler</author><pubDate>Thu, 05 Oct 2017 16:30:57 +0800</pubDate><description><![CDATA[Tensorflow 常见错误及解析Attempting to use uninitialized value说明没有初始化变量，可以加入:        sess.run(tf.global_variables_initializer())
        sess.run(tf.local_variables_initializer())Table not initialized.在使用feat]]></description><category></category></item><item><title><![CDATA[第九课 tensorflow 特征工程: feature_column]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78161748</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78161748</guid><author>cjopengler</author><pubDate>Thu, 05 Oct 2017 16:29:10 +0800</pubDate><description><![CDATA[特征工程: feature_column在使用很多模型的时候，都需要对输入的数据进行必要的特征工程处理。最典型的就是:one-hot处理，还有hash分桶等处理。为了方便处理这些特征，tensorflow提供了一些列的特征工程方法来方便使用.公共的importimport tensorflow as tf
from tensorflow.python.estimator.inputs import]]></description><category></category></item><item><title><![CDATA[第八课 tensorflow numpy_input_fn 以及队列性质]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/78150650</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/78150650</guid><author>cjopengler</author><pubDate>Mon, 02 Oct 2017 12:58:38 +0800</pubDate><description><![CDATA[numpy_input_fn 以及队列性质该函数的作用是从numpy的输入数据中，产生读取的featrues和labels数据。这样当我们在使用numpy的数据作为输入的时候就很方便。对于所有的input来说，都是要建立队列来进行读入，所以对于队列的处理就会比较麻烦，而numpy_input的数据将这些对队列的输入封装在一起方便了我们使用.import tensorflow as tf
import]]></description><category></category></item><item><title><![CDATA[第七课 Tensorflow Cifar10 Eval]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/77948942</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/77948942</guid><author>cjopengler</author><pubDate>Tue, 12 Sep 2017 19:53:40 +0800</pubDate><description><![CDATA[Cifar10的评估# coding:utf-8
"""
评估准确率步骤就是:
1. 载入测试数据集
2. 使用先前构建的模型, 对输入的数据产出预测结果
3. 计算预测结果的正确比率
4. 载入模型计算的参数，运行模型
"""from abc import ABCMeta
from abc import abstractmethod
import tensorflow as tf
import m]]></description><category></category></item><item><title><![CDATA[第六课 Tensorflow Cifar10 CNN 卷积神经网络]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/77948926</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/77948926</guid><author>cjopengler</author><pubDate>Tue, 12 Sep 2017 19:51:41 +0800</pubDate><description><![CDATA[对于train来说，基本就是几个步骤:
读取数据
构建训练网络
训练
# coding:utf-8
"""
训练
"""from abc import ABCMeta
from abc import abstractmethod
class ITrain(object):    __metaclass__ = ABCMeta    @abstractmethod
    def train(self]]></description><category></category></item><item><title><![CDATA[第五课 Tensorflow TFRecord读取数据]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/77675708</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/77675708</guid><author>cjopengler</author><pubDate>Tue, 29 Aug 2017 11:14:35 +0800</pubDate><description><![CDATA[虽然，可以使用常用的类型，但是使用tfrecord更好。
protobuf的格式传输更快
结构统一。相当于屏蔽了底层的数据结构。
import tensorflow as tf
import numpy as np
from IPython.display import display, HTML
import matplotlib.pyplot as plt
import pandas as pd]]></description><category></category></item><item><title><![CDATA[第四课 Tensorflow并发读取数据]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/77675680</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/77675680</guid><author>cjopengler</author><pubDate>Tue, 29 Aug 2017 11:13:01 +0800</pubDate><description><![CDATA[并发读取数据能够大大加速数据的读取速度。import tensorflow as tf
import numpy as np
from IPython.display import display, HTML
import matplotlib.pyplot as pltplt.rcParams["figure.figsize"] = (20,10)filenames = tf.train.matc]]></description><category></category></item><item><title><![CDATA[第三课 Tensorflow实现人工神经网络]]></title><link>https://blog.youkuaiyun.com/cjopengler/article/details/77621553</link><guid>https://blog.youkuaiyun.com/cjopengler/article/details/77621553</guid><author>cjopengler</author><pubDate>Sun, 27 Aug 2017 17:12:05 +0800</pubDate><description><![CDATA[Tesnroflow人工神经网络import numpy as np
import tensorflow as tf
from tensorflow.examples.tutorials import mnist
from IPython.display import display, HTML
import matplotlib.pyplot as plt
读取数据集mnist_data = mn]]></description><category></category></item></channel></rss>