Question

所以我基本上是在分析调查数据集。数据集如下所示：

   Respondent       Country           HaveWorkedLanguage
0     1             United States     Swift
1     2             United Kingdom    JavaScript; Python; Ruby; SQL
2     3             United Kingdom    Java; PHP; Python
3     4             United States     Matlab; Python; R; SQL
4     5             Switzerland       NaN
5     6             New Zealand       JavaScript; PHP; Rust

正如您所看到的， HaveWorkedLanguage 列在每个单元格中都包含单个值或多个值的实例。我想做的是分析每个国家最着名的语言。为此我首先执行了这样的组合：

stu=students.groupby(['Country','HaveWorkedLanguage'])['Respondent'].count().reset_index()
stu.columns=[['Country','Known_Languages','Count']]

我得到了这样的数据框：

    Country         Known_Languages                             Count
0   Afghanistan     Assembly; C; C++; Hack; Java; JavaScript    1
1   Afghanistan     C                                           1
2   Albania         C#; Java; Python; SQL                       1
3   Albania         C++; C#; Java; JavaScript; PHP              1
4   Albania         C++; C#; JavaScript; SQL                    1
5   Albania         C++; Java; JavaScript; PHP; SQL             2

我实际上想要一个显示国家/地区和每种语言数量的数据框，以便最高计数显示最着名的语言。数据框应该是这样的：

      Country           Known_Languages     Count
0     United States    Java                 100
1     United States    Python               80

之前我能够使用以下代码找到整体着名语言：

for i in ['C','C++','C#','Java','Python','R','JavaScript']:
    print(i,':',survey['HaveWorkedLanguage'].apply(lambda x: i in str(x).split('; ')).value_counts()[1])

输出结果为：

C : 6974
C++ : 8155
C# : 12476
Java : 14524
Python : 11704
R : 1634
JavaScript : 22875

但是现在我想把国家与它联系起来。我该怎么做？

Answer 1

hwl = students.HaveWorkedLanguage
cty = students.Country
stu = hwl.str.get_dummies('; ').groupby(cty).sum()
pd.concat(
    [stu.idxmax(1), stu.max(1)],
    axis=1, keys=['Lang', 'Count']
)

                      Lang  Count
Country                          
New Zealand     JavaScript      1
Switzerland           Java      0
United Kingdom      Python      2
United States       Matlab      1

<强> PROJECT / 杀
numpy技术

mask = students.HaveWorkedLanguage.notnull().values
fc, uc = pd.factorize(students.Country.values.astype(str))
hwl = students.HaveWorkedLanguage.values.astype(str)
lol = np.core.defchararray.split(hwl, '; ')
lol[np.flatnonzero(~mask)] = [[]]
i = fc.repeat([len(l) for l in lol])
j, ul = pd.factorize(np.concatenate(lol))
n = uc.size
m = ul.size
counts = np.bincount(i * m + j, minlength=n * m).reshape(n, m)
x = counts.argmax(1)
pd.DataFrame(
    np.column_stack([ul[x], counts[np.arange(n), x]]),
    uc, ['Lang', 'Count'])

                      Lang Count
United States        Swift     1
United Kingdom      Python     2
Switzerland          Swift     0
New Zealand     JavaScript     1

计时

%%timeit
hwl = students.HaveWorkedLanguage
cty = students.Country
stu = hwl.str.get_dummies('; ').groupby(cty).sum()
pd.concat(
    [stu.idxmax(1), stu.max(1)],
    axis=1, keys=['Lang', 'Count']
)
100 loops, best of 3: 3.22 ms per loop

%%timeit
mask = students.HaveWorkedLanguage.notnull().values
fc, uc = pd.factorize(students.Country.values.astype(str))
hwl = students.HaveWorkedLanguage.values.astype(str)
lol = np.core.defchararray.split(hwl, '; ')
lol[np.flatnonzero(~mask)] = [[]]
i = fc.repeat([len(l) for l in lol])
j, ul = pd.factorize(np.concatenate(lol))
n = uc.size
m = ul.size
counts = np.bincount(i * m + j, minlength=n * m).reshape(n, m)
x = counts.argmax(1)
pd.DataFrame(np.column_stack([ul[x], counts[np.arange(n), x]]), uc, ['Lang', 'Count'])
1000 loops, best of 3: 570 µs per loop

Answer 2

我已经做了很多步骤，所以也许有人有更多的pythonic解决方案：

df = pd.DataFrame({"Country":["UK", "UK", "UK", "USA", "USA", "USA"], "Languages":["Python" , "Python, PHP, Java", "Java", "Python", "Java", "Python, Javascript"]})
df

    Country Languages
0   UK  Python
1   UK  Python, PHP, Java
2   UK  Java
3   USA Python
4   USA Java
5   USA Python, Javascript

df2 = df.Languages.apply(lambda row: pd.Series(row.split(","))).copy() # split the column
df3 = pd.get_dummies(df2, prefix_sep="", prefix="") # get dummies
df3


    Java    Python  Javascript  PHP Java
0   0   1   0   0   0
1   0   1   0   1   1
2   1   0   0   0   0
3   0   1   0   0   0
4   1   0   0   0   0
5   0   1   1   0   0

df4 = pd.merge(df[["Country"]], df3,  left_index=True, right_index=True)
df4

    Country Java    Python  Javascript  PHP Java
0   UK  0   1   0   0   0
1   UK  0   1   0   1   1
2   UK  1   0   0   0   0
3   USA 0   1   0   0   0
4   USA 1   0   0   0   0
5   USA 0   1   1   0   0

df5 = df4.groupby("Country").sum().reset_index().copy() # sum it
df5

    Country Java    Python  Javascript  PHP Java
0   UK  1   2   0   1   1
1   USA 1   2   1   0   0

df6 = pd.melt(df5, id_vars=["Country"], var_name="Language", value_name="Value") # columns to rows
df6

    Country Language    Value
0   UK  Java    1
1   USA Java    1
2   UK  Python  2
3   USA Python  2
4   UK  Javascript  0
5   USA Javascript  1
6   UK  PHP 1
7   USA PHP 0
8   UK  Java    1
9   USA Java    0

df7 = df6.sort_values(by=["Country", "Value"], ascending=False) # sort
df7


    Country Language    Value
3   USA Python  2
1   USA Java    1
5   USA Javascript  1
7   USA PHP 0
9   USA Java    0
2   UK  Python  2
0   UK  Java    1
6   UK  PHP 1
8   UK  Java    1
4   UK  Javascript  0

如何计算具有多个逗号分隔值的列中单词的实例数？

2 个答案: