# 对比两家电影公司的电影发行情况
# 对电影公司的数据进行整理
moviesdf['prodcompanies']=moviesdf['production_companies'].apply(json.loads)
moviesdf['prodcompany']=moviesdf['prodcompanies'].apply(decode) # 用到上面定义的函数
moviesdf.head()

moviesdf['Universal']=moviesdf['prodcompany'].str.contains('Universal Pictures').apply(lambda x:i if x else 0)
moviesdf['Paramount']=moviesdf['prodcompany'].str.contains('Paramount Pictures').apply(lambda x:i if x else 0)
moviesdf.tail()

a=moviesdf['Universal'].sum()
b=moviesdf['Paramount'].sum()
dict_companies={'Universal':a,'Paramount':b}
dict_companies
companies_number=pd.Series(dict_companies)
companies_number.plot(kind='pie',label='',autopct='%11.1f%%')

company=moviesdf[['Universal','Paramount']]
company.index=moviesdf['releaseyear']
companydf=company.groupby('releaseyear').sum()
companydf.head()
plt.figure(figsize=(10,7))
plt.plot(companydf,label=companydf.columns)

# 通过keywords字段，分割改编电影与原创电影的对比
# 分析：在keywords字段中，如果存在based on novel 则说明电影改编自小说
moviesdf['keywds']=moviesdf['keywords'].apply(json.loads)
moviesdf['keywds2']=moviesdf['keywds'].apply(decode)
# 标注是否为基于小说改编的电影
a='based on novel'
moviesdf['if_original']=moviesdf['keywds2'].str.contains(a).apply(lambda x:'no original' if x else 'original')
moviesdf['if_original'].value_counts()
# moviesdf.head(2)
original_profit=moviesdf[['if_original','budget','revenue','profit']]
original_profit_mean=original_profit.groupby(by='if_original').mean()
original_profit_mean.head()
plt.figure(figsize=(12,9))
original_profit_mean.plot(kind='bar')
plt.title('原创电影与改编电影的预算,收入,利润比较')
plt.xlabel('原创与改编')
plt.ylabel('现金流')
plt.xticks(rotation=0)

# 电影的runtime与票房，评分之间的关系
moviesdf.plot(kind='scatter',x='runtime',y='revenue',figsize=(10,7))
plt.title('电影时长与票房的关系',fontsize=15)
plt.xlabel('时长（分钟）',fontsize=15)
plt.ylabel('票房收入',fontsize=15)

# 电影的时长与评分之间的关系
moviesdf.plot(kind='scatter',x='runtime',y='vote_average',figsize=(10,7))
plt.title('电影时长与评分之间的关系',fontsize=15)
plt.xlabel('时长（分钟）',fontsize=15)
plt.ylabel('评分',fontsize=15)

# 电影关键字词频计算和词云生成
keywords_list=[]
for i in moviesdf['keywds2']:
    keywords_list.append(i)
#     keywords_list=list(keywords_list)
lis=''.join(keywords_list)
lis.replace('\'s','')
lis

# 根据电影关键词生成词频计算
import jieba
words=jieba.lcut(lis)
counts={}
mytext_list=[]
for word in words:
    if len(word)==1: # 单个分词不需要
        continue 
    else:
        counts[word]=counts.get(word,0)+1
        mytext_list.append(word.replace(" ",""))
cloud_text=','.join(mytext_list)
# print(cloud_text)
items=list(counts.items())
items.sort(key=lambda x:x[1],reverse=True)
for i in range(100):
    word,countsnt=items[i]
    print("{0:<10}{1:>5}".format(word,count))

from PIL import Image  #需要下载安装PIL包
import numpy as np
from wordcloud import WordCloud, ImageColorGenerator
from matplotlib import pyplot as plt
import pandas as pd 

# 简单生成词云
wc=WordCloud(
    background_color='black',
    max_words=2000,
    max_font_size=100,
    min_font_size=10,
    random_state=12
)
wc.generate(lis)
plt.figure(figsize=(12,9))
plt.imshow(wc)
plt.axis('off')
plt.show()

# 安装pkuseg
!pip install pkuseg
#调用北大算法和jieba算法得到的结果对于英文来说基本相同，但对于中文来说pkuseg会更加准确

#根据电影关键词生成词频计算　北大算法pkuseg
# 根据电影关键词生成词频计算
import pkuseg
seg=pkuseg.pkuseg()
words=seg.cut(lis) # 与jieba的区别主要是调用的算法不同
counts={}
mytext_list=[]
for word in words:
    if len(word)==1: # 单个分词不需要
        continue 
    else:
        counts[word]=counts.get(word,0)+1
        mytext_list.append(word.replace(" ",""))
cloud_text=','.join(mytext_list)
# print(cloud_text)
items=list(counts.items())
items.sort(key=lambda x:x[1],reverse=True)
for i in range(100):
    word,count=items[i]
    print("{0:<20}{1:>5}".format(word,count))

# 分析演员的关联度（社会网络计算） 
moviesdf['actorsls']
actors_list=[]
actorsforcut=[]
# 将每部电影的演员放在列表当中
for i in moviesdf['actorsls']:
    actors_list.append(i.split(','))
# 将演员列表中的对象放置到为分割而设置的另一个列表
for k in range(0,len(actors_list)):
    for j in range(len(actors_list[k])):
        actorsforcut.append(actors_list[k][j])
actors_list=''.join(actorsforcut)
actors_list

wc=WordCloud(
    background_color='black',
    max_words=2000,
    max_font_size=100,
    min_font_size=10,
    random_state=12
)
wc.generate(actors_list)
plt.figure(figsize=(12,9))
plt.imshow(wc)
plt.axis('off')
plt.show()

# 改进方案：
# 分析演员的关联度（社会网络计算） 
moviesdf['actorsls']
actors_list=[]
actorsforcut=[]
# 将每部电影的演员放在列表当中
for i in moviesdf['actorsls']:
    actors_list.append(i.split(','))
# 将演员列表中的对象放置到为分割而设置的另一个列表
for k in range(0,len(actors_list)):
    for j in range(len(actors_list[k])):
        actorsforcut.append("'"+actors_list[k][j]+"'") # 在这里对每个人的姓名前后都加上单引号
actors_list=','.join(actorsforcut)
actors_list[0:100]

# 如何将演员中间的空格去除，并合并为一个整体
movieactorsdfv2=moviesdf['actorsls'].apply(lambda x:''.join(x.split())).replace()
movieactorsdfv2.head(2)
movieactorsdfv2.to_csv('actors.txt',index=False,header=False)

# 读取数据并加载字典
with open('actors.txt',encoding='utf-8-sig') as f:
    actorstxt=f.readlines()
# 此处的actors_forcut是手工查找替换单引号等完成的，在进行DF转换中可以直接处理
jieba.load_userdict("actors_forcut")

actorsforcut
actorsdf=pd.DataFrame(actorsforcut).drop_duplicates()
actorsdf[0]=actorsdf[0].apply(lambda x:''.join(x.split())).replace() # 将中间的空格删除
# 此时拿到了完整的人物列表
actorsdf[0]=actorsdf[0].apply(lambda x:''.join(x.split('\''))).replace() # 将姓名两边的单引号去除
actorsdf['']='nr'
actorsdf.to_excel('actorsdf.xlsx',index=False)
actorsdf.head()

movieactorsdfv2=pd.DataFrame(moviesdf['actorsls'])
movieactorsdfv2.head(5)