python pandas tensorflow使用总结

################## pd tf 相关使用技巧 ##################
python 函数只能都放在一个包里。。。
python 的with 相当于 golang 的 defer
python 包在导入时不能互相嵌套, test1 import test2 test2 import test1, 一旦如此， 就会无法调用；

## 列表、字典判断 IO 异常处理 ##

#列表、字典不为空的判断:
if l == []:
if m == {}:

#字典中查找是否存在key:
if 'key' in test.keys():

#LinkedMap
from collections import OrderedDict, defaultdict

#IO
with open("./sql.txt",'w') as fw:
l = map(lambda x:x+" ",l)
fw.writelines(list(l))

with open("./sql.txt",'r') as fr:
lines = fr.readlines()
print(lines)

os.system(ex) # 执行bash命令

#获取时间
datetime.today().strftime('%Y%m%d')

#获取文件地址:
path = os.getcwd()
file_path = os.path.join(path,'prod.cfg')

#字符串切分: 直接用[]
if line.find("#")!=-1:
line=line[0:line.find('#')]

#异常处理：
try:
except Exception as e:
raise e

## 匿名函数 ##
# map sorted filter

from typing import Any, Tuple, Iterator

# map
l = [1, 2, 3, 4, 5]
t1: Iterator[Tuple[Any, int]] = map(lambda x: (x, 1), l)

# sorted
m = dict({"a": 1, "b": 0})

t = sorted(m.items(), key=lambda d: d[1], reverse=False) # 按照value排序， 升序

# filter
f = list(filter(lambda x: x[0].find('b') == -1, m))

## 枚举迭代删除 ##

# for enumerate

some_string = "wtf"
some_dict = {}
for i, some_dict[i] in enumerate(some_string):
pass

print(some_dict)

# for zip
index = [1,2,3]
words = ['a','b','c']
for i, w in zip(index, words):
pass

# 反转列表
for i in reversed(index):
pass

for i in index[::-1]:
pass

# all any 判断列表中的所有值是否与条件匹配;

r = any(i != 1 for i in index)
print(r)

## 在迭代时 删除原表 需要借助副本 ##

list_3 = [1,2,3,4]
for idx, item in enumerate(list_3[:]):
list_3.remove(item)
print(list_3)

list_3 = [1,2,3,4]
list_temp = list_3.copy()
for idx, item in enumerate(list_3[:]):
list_3.remove(item)
print(list_3)

## Pandas操作 ##

import pandas as pd

data = {'a':[1,2,3],
'c':[4,5,6],
'b':[7,8,9]
}

# 创建DF
frame = pd.DataFrame(data,index=['1','2','3'])

# group by
d: Union[Union[SeriesGroupBy, DataFrameGroupBy], Any] = df.groupby("vin")

for key, group_data in d:

# key, group_data 是一个list[DataFrame]

for i in range(0, len(group_data)):
group_data.iloc[i]["mileage"] #遍历iloc第i行df, 取出mileage列;
pass

# 如果想改变值， 无法在iloc切片上直接改变， 需要复制一份加到list中;

# 读取多列:
result = df[["task_name","task_name_en"]]
# 读取多行:
result.iloc[[0,1,2,3]]

# df拼接
pd.DataFrame([c1,c2]), pd.concat([p1,p2])时, 首先要保证各df的列数相同，如果还是不行:
t = {"task_name":result["task_name"].to_list(),"table_name":result["table_name"].to_list(),"content_crt":l_crt, "content_ist":l_ist}
f = pd.DataFrame(t)

#numpy:
np.random.randint(-1,1,size=(5,5)) | np.random.uniform #不重复的
numpy.take(m,1) # 取出每一行的第m列

## 通过map 改变pd字段的值；

gender_map = {'F':0, 'M':1}
users['Gender'] = users['Gender'].map(gender_map)

age_map = {val:ii for ii,val in enumerate(set(users['Age'])) } # 用字典做替换: { 原始值 : 新值 }
users['Age'] = users['Age'].map(age_map)

pattern = re.compile(r'^(.*)((d+))$') # 反斜杠+( 转义( Toy Story (1995) -> Toy Story
title_map = {val: pattern.match(val).group(1) for ii,val in enumerate(set(movies['Title'])) }

#### tensorflow 测试 ####

1. 组件
使用graph -> 表示计算任务
使用session.context -> 执行任务
使用tensor  -> 表示数据                  每个Tensor 是多维数组[batch, height, width, channels] .ndarray
使用Variable -> 维护状态
使用feed fetch -> 赋值与获取数据


2. 组件使用
-> 一个常量为一个节点 op, 例如创建两个节点
matrix = tf.constant([[2., 1.], [2., 1.]])     # 常量不需要初始化， 而变量需要
product = tf.matmul(matrix, matrix)


-> 执行计算任务 使用with自动释放资源, 代替sess.close()
with tf.Session() as sess:
    #with tf.device("/gpu:1"):
    rs = sess.run(product)
    print(rs)

-> 变量需要初始化， 使用InteractiveSession()交互环境， Tensor.eval() 和 Operation.run() 方法代替 Session.run()
sess = tf.InteractiveSession()      #不需要使用sess.run() 而是直接 op.run() 即可初始化变量, 它能让你在运行图的时候，插入一些计算图

x = tf.Variable([[1.0, 2.0], [2.0, 4.0]])
x.initializer.run()                 # 如果不使用交互模式， 需要tf.initialize_all_variables()

sub = tf.subtract(x, matrix)
print(sub.eval())


-> 改变一个节点op的状态， 即计数器
state = tf.Variable(0, name="count")

one = tf.constant(1)
add_op = tf.add(state, one)
update = tf.assign(state, add_op)       #更新节点状态

->  Fetch 与 Feed

input_1 = tf.placeholder(tf.dtypes.float32)         # 通过 run(feed_dict:)  feed
input_2 = tf.placeholder(tf.dtypes.float32)
out_1 = tf.add(input_1, input_2)                    # 通过run fetch
out_2 = tf.subtract(input_1, input_2)


with tf.Session() as sess:
    o1, o2 = sess.run([out_1, out_2], feed_dict={input_1: [7.], input_2: [5.]})
    print(o1, o2)

4. 可视化与保存

# # 训练可视化
# summary_op = tf.merge_all_summaries()
# summary_writer = tf.train.SummaryWriter("train_dir",
#                                         graph_def=sess.graph_def)
#
# summary_str = sess.run(summary_op, feed_dict=feed_dict)
# summary_writer.add_summary(summary_str, step)
#
# # 保存参数
# saver = tf.train.Saver()
# saver.save(sess, FLAGS.train_dir, global_step=step)
#
# saver.restore(sess, FLAGS.train_dir)


# 启动TensorBoard　
# python tensorflow/tensorboard/tensorboard.py --logdir=path/to/log-directory
# tensorboard --logdir=/path/to/log-directory




# ################# 使用


Tensor.get_shape()[1]

# 它两必须一起
sess.run(tf.initialize_all_variables())
d1 = sess.run(h_gen, feed_dict={x: x_data, z:z_d})

init_op = tf.initialize_all_variables()

with tf.Session() as sess:
    # 在任务中初始化变量
    sess.run(init_op)

    for _ in range(3):
        sess.run(update)
        print(state.eval())

import tensorflow as tf
import numpy as np

uid_max = 500
batch_size = 10
embed_dim = 32
filter_num = 8

feature_num = 20

data = np.zeros((batch_size,feature_num))

uid_data = np.reshape(data, [batch_size, feature_num])

sess = tf.InteractiveSession()

uid = tf.placeholder(tf.int32, [None, feature_num], name="uid")

uid_embed_matrix = tf.Variable(tf.random_uniform([uid_max, 32], -1, 1),
name="uid_embed_matrix")
# 根据指定用户ID找到他对应的嵌入层
uid_embed_layer = tf.nn.embedding_lookup(uid_embed_matrix, uid,
name="uid_embed_layer")

new_layer = tf.reduce_sum(uid_embed_layer, axis=1, keep_dims=True) # 按列加和， 维度保持不变；
new_layer = tf.expand_dims(uid_embed_layer, -1) # 对卷积而言特殊使用的， 转为 (batch_size, feature_num, 32, 1)

filter_weights = tf.Variable(tf.truncated_normal([2, embed_dim, 1, filter_num],stddev=0.1),name = "filter_weights") # 卷积部分
filter_bias = tf.Variable(tf.constant(0.1, shape=[filter_num]), name="filter_bias")

conv_layer = tf.nn.conv2d(new_layer, filter_weights, [1,1,1,1], padding="VALID", name="conv_layer")
relu_layer = tf.nn.relu(tf.nn.bias_add(conv_layer,filter_bias), name ="relu_layer")
maxpool_layer = tf.nn.max_pool(relu_layer, [1,15 - 2 + 1 ,1,1], [1,1,1,1], padding="VALID", name="maxpool_layer")

sess.run(tf.initialize_all_variables())

feed_dict = {uid:uid_data}

layer = uid_embed_layer.eval(feed_dict)

print(layer.shape)