《Python for Data Analysis》一书由Wes Mckinney所著，中文译名是《利用Python进行数据分析》。这里记录一下学习过程，其中有些方法和书中不同，是按自己比较熟悉的方式实现的。
第二个实例：MovieLens 1M Data Set
简介： GroupLens Research提供了从MovieLens用户那里收集来的一系列对90年代电影评分的数据。
数据地址：http://files.grouplens.org/datasets/movielens/ml-1m.zip
准备工作：导入pandas和matplotlib
import pandas as pd import matplotlib.pyplot as plt fig,ax=plt.subplots()
压缩包里有三个.dat文件，分别是movies, users, ratings。这几个文件可以用pandas的read_table()方法读入并变为DataFrame格式，用names参数设置各个表的列名。
movies=pd.read_table(r"...\movies.dat", sep='::', engine='python', names=["movieId", "title", "genre"]) users=pd.read_table(r"...\users.dat", sep='::', engine='python', names=["userId", "gender", "age", "occupation", "zip"]) ratings=pd.read_table(r"...\ratings.dat", sep='::', engine='python', names=["userId", "movieId", "rating", "timestamp"])
接下来把这三张表合并在一起，以便于分析。其中movies和ratings先通过movieId列进行连接，然后合并的表再与users通过userId列进行连接。
data=pd.merge(pd.merge(movies, ratings, on="movieId", how="inner"), users, on="userId", how="inner")
合并的表前5行显示如下：
movieId title \ 0 1 Toy Story (1995) 1 48 Pocahontas (1995) 2 150 Apollo 13 (1995) 3 260 Star Wars: Episode IV - A New Hope (1977) 4 527 Schindler's List (1993) genre userId rating timestamp gender \ 0 Animation|Children's|Comedy 1 5 978824268 F 1 Animation|Children's|Musical|Romance 1 5 978824351 F 2 Drama 1 5 978301777 F 3 Action|Adventure|Fantasy|Sci-Fi 1 4 978300760 F 4 Drama|War 1 5 978824195 F age occupation zip 0 1 10 48067 1 1 10 48067 2 1 10 48067 3 1 10 48067 4 1 10 48067
上面可以看到，age这一列有明显的异常（1岁？），因此这里把data中age小于18岁和大于100岁的人去除。
data=data[(data["age"]>=18) & (data["age"]<=100)]
我们来看一下，按性别分组，对各部电影的平均评分是多少：
by_gender_movie_rating=pd.pivot_table(data, values="rating", index="title", columns="gender", aggfunc="mean")
这里用透视表展示了男女分别对各部电影的平均评分：
gender F M title $1,000,000 Duck (1971) 3.375000 2.761905 'Night Mother (1986) 3.400000 3.424242 'Til There Was You (1997) 2.694444 2.571429 'burbs, The (1989) 2.793478 2.947368 ...And Justice for All (1979) 3.828571 3.693252 1-900 (1994) 2.000000 3.000000 10 Things I Hate About You (1999) 3.593137 3.303855 101 Dalmatians (1961) 3.789474 3.512535 101 Dalmatians (1996) 3.210526 2.928934 12 Angry Men (1957) 4.229008 4.318376
然而，我们考虑到如果一部电影打分的人太少，那么此评分就不会太准确，该电影就不能作为取样。因此，我们要对每部电影打分的人数进行统计，并把评分人数超过250的电影筛选出来。
movie_counts=data.groupby('title')['title'].count() movies_select=movie_counts.index[movie_counts.values>=250]
然后，我们把上面的透视表按照选出的电影movies_select进行筛选，选出所有符合条件的行：
by_gender_movie_rating=by_gender_movie_rating.loc[movies_select]
我们再来看看现在透视表变成了什么样：
gender F M title 'burbs, The (1989) 2.793478 2.947368 10 Things I Hate About You (1999) 3.593137 3.303855 101 Dalmatians (1961) 3.789474 3.512535 101 Dalmatians (1996) 3.210526 2.928934 12 Angry Men (1957) 4.229008 4.318376 13th Warrior, The (1999) 3.084746 3.172185 2 Days in the Valley (1996) 3.477273 3.246862 20,000 Leagues Under the Sea (1954) 3.648936 3.723404 2001: A Space Odyssey (1968) 3.829341 4.125931 2010 (1984) 3.456522 3.418269
现在，如果我们想知道女性评分最高的10部电影分别是什么，那么我们可以对F列的值进行排序：
top_female_rating=by_gender_movie_rating.sort_values(by='F', ascending=False)
以下是结果：
gender F M title Close Shave, A (1995) 4.672619 4.479121 Wrong Trousers, The (1993) 4.611607 4.485390 Wallace & Gromit: The Best of Aardman Animation... 4.587629 4.413043 Grand Day Out, A (1992) 4.581967 4.288820 Sunset Blvd. (a.k.a. Sunset Boulevard) (1950) 4.575221 4.476744 Schindler's List (1993) 4.563333 4.493325 To Kill a Mockingbird (1962) 4.539792 4.395387 Shawshank Redemption, The (1994) 4.539088 4.560944 Creature Comforts (1990) 4.514286 4.287958 Usual Suspects, The (1995) 4.512255 4.520864
现在，我们想看一下男女评分差异最大的10部电影分别是什么。首先，给透视表增加差别列-diff，然后再对diff列的值进行排序。
import numpy as np by_gender_movie_rating["diff"]=np.abs(by_gender_movie_rating["F"]-by_gender_movie_rating["M"]) top_diff_rating=by_gender_movie_rating.sort_values(by='diff', ascending=False)
来看一下top_diff_rating的前10行：
gender F M diff title Dirty Dancing (1987) 3.762590 2.961929 0.800661 Good, The Bad and The Ugly, The (1966) 3.484536 4.223776 0.739240 Jumpin' Jack Flash (1986) 3.269231 2.582707 0.686524 Kentucky Fried Movie, The (1977) 2.875000 3.555970 0.680970 Dumb & Dumber (1994) 2.700000 3.318275 0.618275 Hidden, The (1987) 3.137931 3.744094 0.606163 Cable Guy, The (1996) 2.280488 2.878472 0.597984 Grease (1978) 3.958955 3.376673 0.582282 Rocky III (1982) 2.361702 2.939828 0.578126 Evil Dead II (Dead By Dawn) (1987) 3.328767 3.900000 0.571233
如果想知道不论男女，所有观众评分差异最大的10部电影，那么我们先计算出总评分的标准差，再提取评分人数超过250的电影，最后按标准差进行排序。
movie_rating_std=data.groupby('title')['rating'].std() movie_rating_std=movie_rating_std.loc[movies_select] top_rating_std=movie_rating_std.sort_values(ascending=False)
结果如下：
title Dumb & Dumber (1994) 1.324767 Blair Witch Project, The (1999) 1.319496 Natural Born Killers (1994) 1.305525 Tank Girl (1995) 1.278513 Rocky Horror Picture Show, The (1975) 1.259985 Eyes Wide Shut (1999) 1.254972 Fear and Loathing in Las Vegas (1998) 1.247835 Evita (1996) 1.247072 Hellraiser (1987) 1.243238 South Park: Bigger, Longer and Uncut (1999) 1.237987 Name: rating, dtype: float64
至此，书中的分析已全部结束。以下是我自己增加的一些分析内容：
如果我们想知道总评分最高的10部电影，男女评分之间有没有很大的差异，那么我们先在上面的透视表by_gender_movie_rating里增加一个总评分栏，然后按照总评分进行排序。
movie_rating=data.groupby('title')['rating'].mean() #先把总平均评分算出来 movie_rating=movie_rating.loc[movies_select] #摘选评分人数超过250的电影 by_gender_movie_rating['total']=movie_rating.values #在透视表中增加总平均评分一列 top_rating=by_gender_movie_rating.sort_values(by='total', ascending=False) #按总平均评分排序
top_rating的前10行如下：
gender F M \ title Seven Samurai (The Magnificent Seven) (Shichini... 4.471154 4.580392 Shawshank Redemption, The (1994) 4.539088 4.560944 Close Shave, A (1995) 4.672619 4.479121 Godfather, The (1972) 4.319829 4.583186 Wrong Trousers, The (1993) 4.611607 4.485390 Usual Suspects, The (1995) 4.512255 4.520864 Schindler's List (1993) 4.563333 4.493325 Sunset Blvd. (a.k.a. Sunset Boulevard) (1950) 4.575221 4.476744 Raiders of the Lost Ark (1981) 4.341727 4.529474 Rear Window (1954) 4.475524 4.482480 gender diff total title Seven Samurai (The Magnificent Seven) (Shichini... 0.109238 4.561889 Shawshank Redemption, The (1994) 0.021857 4.554791 Close Shave, A (1995) 0.193498 4.531300 Godfather, The (1972) 0.263357 4.526267 Wrong Trousers, The (1993) 0.126218 4.519048 Usual Suspects, The (1995) 0.008609 4.518857 Schindler's List (1993) 0.070008 4.512011 Sunset Blvd. (a.k.a. Sunset Boulevard) (1950) 0.098477 4.501094 Raiders of the Lost Ark (1981) 0.187748 4.486675 Rear Window (1954) 0.006955 4.480545
用柱形图画出来进行比较：
ax.bar(range(10), top_rating['F'][:10], width=-0.3, label='Female', align='edge') ax.bar([i+0.3 for i in range(10)], top_rating['M'][:10], width=-0.3, label='Male', align='edge') ax.set_xticks(range(10)) ax.set_ylim(4,5) ax.set_xticklabels(top_rating[:10].index.values, rotation=90) ax.legend() plt.show()
可以看到，男性对教父的评分比女性要高很多。
现在，让我们再来看看各个年龄段最喜欢的10部电影是什么。
首先，对年龄进行分组，并在data数据里添加年龄分组这一列：
age_range=pd.cut(data['age'], 3, labels=['Young', 'Middle', 'Old']) data['age_range']=age_range
data的前10行现在如下：
movieId title \ 53 1 Toy Story (1995) 54 17 Sense and Sensibility (1995) 55 34 Babe (1995) 56 48 Pocahontas (1995) 57 199 Umbrellas of Cherbourg, The (Parapluies de Che... 58 266 Legends of the Fall (1994) 59 296 Pulp Fiction (1994) 60 364 Lion King, The (1994) 61 368 Maverick (1994) 62 377 Speed (1994) genre userId rating timestamp gender \ 53 Animation|Children's|Comedy 6 4 978237008 F 54 Drama|Romance 6 4 978236383 F 55 Children's|Comedy|Drama 6 4 978237444 F 56 Animation|Children's|Musical|Romance 6 5 978237570 F 57 Drama|Musical 6 5 978237570 F 58 Drama|Romance|War|Western 6 4 978237909 F 59 Crime|Drama 6 2 978237379 F 60 Animation|Children's|Musical 6 4 978237570 F 61 Action|Comedy|Western 6 4 978237909 F 62 Action|Romance|Thriller 6 3 978236383 F age occupation zip age_range 53 50 9 55117 Old 54 50 9 55117 Old 55 50 9 55117 Old 56 50 9 55117 Old 57 50 9 55117 Old 58 50 9 55117 Old 59 50 9 55117 Old 60 50 9 55117 Old 61 50 9 55117 Old 62 50 9 55117 Old
然后，按年龄段作为列，制作透视表：
by_age_movie_rating=pd.pivot_table(data, values="rating", index="title", columns="age_range", aggfunc="mean")
这里透视表by_age_movie_rating展示了各个年龄段的观众对各部电影的平均评分：
age_range Middle Old Young title $1,000,000 Duck (1971) 3.133333 2.600000 3.058824 'Night Mother (1986) 2.904762 3.777778 3.551724 'Til There Was You (1997) 2.900000 2.500000 2.625000 'burbs, The (1989) 2.818182 2.951220 2.912195 ...And Justice for All (1979) 3.657143 3.809524 3.692308 1-900 (1994) NaN 3.000000 2.000000 10 Things I Hate About You (1999) 3.102941 3.476190 3.424125 101 Dalmatians (1961) 3.826087 3.692308 3.488746 101 Dalmatians (1996) 3.279570 3.460317 2.764368 12 Angry Men (1957) 4.358333 4.268156 4.293333
可以看到上面有无效字段，应该是该年龄段没有人对此电影进行评分。因此，我们把无效值变为0。同时，把评分人数超过250的电影筛选出来：
by_age_movie_rating=by_age_movie_rating.fillna(0)
by_age_movie_rating=by_age_movie_rating.loc[movies_select]
然后我们按Young这一列的值来排序，看看年轻人评分最高的10部电影是什么：
top_young_movie_rating=by_age_movie_rating.sort_values(by='Young', ascending=False)
结果如下：
age_range Middle Old \ title Shawshank Redemption, The (1994) 4.487500 4.423690 Usual Suspects, The (1995) 4.390879 4.319231 Seven Samurai (The Magnificent Seven) (Shichini... 4.532895 4.581006 Godfather, The (1972) 4.541935 4.452290 Close Shave, A (1995) 4.450704 4.577465 Star Wars: Episode IV - A New Hope (1977) 4.354633 4.386760 Raiders of the Lost Ark (1981) 4.475538 4.414737 Wrong Trousers, The (1993) 4.443850 4.663866 Rear Window (1954) 4.479245 4.461818 Sunset Blvd. (a.k.a. Sunset Boulevard) (1950) 4.611570 4.432624 age_range Young title Shawshank Redemption, The (1994) 4.617735 Usual Suspects, The (1995) 4.595943 Seven Samurai (The Magnificent Seven) (Shichini... 4.565371 Godfather, The (1972) 4.552921 Close Shave, A (1995) 4.551220 Star Wars: Episode IV - A New Hope (1977) 4.524260 Raiders of the Lost Ark (1981) 4.514109 Wrong Trousers, The (1993) 4.513109 Rear Window (1954) 4.491803 Sunset Blvd. (a.k.a. Sunset Boulevard) (1950) 4.482051
第一名是肖申克的救赎。
用同样的方法，我们可以看到老年组评分最高的电影是Wrong Trousers, The (1993)，中年组评分最高的电影是Sunset Blvd. (a.k.a. Sunset Boulevard) (1950)。