Chapter 2: Essential DataFrame Operations

注意选择多列时，不要用tuple，要用list
movie['actor_1_name', 'actor_2_name', 'actor_3_name', 'director_name'] # 会报错
# 上述等价于 movie[('actor_1_name', 'actor_2_name', 'actor_3_name', 'director_name')]

# 正确的用法是list
cols =[['actor_1_name', 'actor_2_name', 'actor_3_name', 'director_name']]

可以根据列的dtype/列名来选择多列
movie.get_dtype_counts() # 先看看各种dtype的列的个数
movie.select_dtypes(include=['number']).head() # 只看数值列
movie.filter(like='facebook').head() # 列名称包含子串
movie.filter(regex='\d').head() # 正则能match上
movie.filter(items=['actor_1_name', 'asdf']).head() # 跟columns取交集，多写的不会报错（如不存在的'asdf'列）

列名排序最佳实践
* 离散靠左，连续靠右
* 比较重要的列靠左
# 按照最佳实践排好序
disc_core = ['movie_title','title_year', 'content_rating','genres']
disc_people = ['director_name','actor_1_name', 'actor_2_name','actor_3_name']
disc_other = ['color','country','language','plot_keywords','movie_imdb_link']
cont_fb = ['director_facebook_likes','actor_1_facebook_likes','actor_2_facebook_likes',
           'actor_3_facebook_likes', 'cast_total_facebook_likes', 'movie_facebook_likes']
cont_finance = ['budget','gross']
cont_num_reviews = ['num_voted_users','num_user_for_reviews', 'num_critic_for_reviews']
cont_other = ['imdb_score','duration', 'aspect_ratio', 'facenumber_in_poster']

# 拼接成大的list，作为新的columns
new_col_order = disc_core + disc_people + disc_other + \
                    cont_fb + cont_finance + cont_num_reviews + cont_other
# 记得check一下，是否有遗漏
set(movie.columns) == set(new_col_order)

# 通过选择来reorder_columns
movie2 = movie[new_col_order]

pandas 中的统计方法一般默认会skipna，安静地过滤掉了，有点危险，可以强制改变这一行为 `movie.min(skipna=False)`

写单元测试时，可以用到这个包
from pandas.testing import assert_frame_equal
assert_frame_equal(college_ugds_, college_ugds_)
# 注意eq方法不同，返回的是同shape的DataFrame
college_ugds_.eq(.0019).head()