如何按熊猫中的一系列值分组？

Question

我有一个dataframe，我想按分类变量和一系列值对其进行分组。您可能会将其视为类似值的行（集群？）。例如：

df = pd.DataFrame({'symbol' : ['IP', 'IP', 'IP', 'IP', 'IP', 'IP', 'IP'],
                   'serie' : ['A', 'B', 'A', 'B', 'A', 'B', 'B'],
                   'strike' : [10, 10, 12, 13, 12, 13, 14],
                   'last' : [1, 2, 2.5, 3, 4.5, 5, 6],
                   'price' : [11, 11, 11, 11, 11, 11, 11],
                   'type' : ['call', 'put', 'put', 'put', 'call', 'put', 'call']})

如果我使用

grouped = df.groupby(['symbol', 'serie', 'strike'])

我已经解决了部分问题，但我想组合更接近的罢工值，例如 10 和 11、12 和 13 等等。优选在％范围内。

Answer 1

在strike 的 bins 上执行 groupy()

使用pd.cut 创建罢工数据箱，然后按该信息分组：

# Create DataFrame
df = pd.DataFrame({
    'symbol' : ['IP', 'IP', 'IP', 'IP', 'IP', 'IP', 'IP'],
    'serie' : ['A', 'B', 'A', 'B', 'A', 'B', 'B'],
    'strike' : [10, 10, 12, 13, 12, 13, 14],
    'last' : [1, 2, 2.5, 3, 4.5, 5, 6],
    'price' : [11, 11, 11, 11, 11, 11, 11],
    'type' : ['call', 'put', 'put', 'put', 'call', 'put', 'call']
})
# Create Bins (example three bins across data)
df['strikebins'] = pd.cut(df['strike'], bins=3)

print 'Binned DataFrame:'
print df
print

# Group these DataFrame
grouped = df.groupby(['symbol', 'serie', 'strikebins'])

# Do something with groups for example
gp_sum = grouped.sum()

print 'Grouped Sum (for example):'
print gp_sum
print

---

Binned DataFrame:
   last  price serie  strike symbol  type        strikebins
0   1.0     11     A      10     IP  call   (9.996, 11.333]
1   2.0     11     B      10     IP   put   (9.996, 11.333]
2   2.5     11     A      12     IP   put  (11.333, 12.667]
3   3.0     11     B      13     IP   put      (12.667, 14]
4   4.5     11     A      12     IP  call  (11.333, 12.667]
5   5.0     11     B      13     IP   put      (12.667, 14]
6   6.0     11     B      14     IP  call      (12.667, 14]

Grouped Sum (for example):
                               last  price  strike
symbol serie strikebins                           
IP     A     (9.996, 11.333]      1     11      10
             (11.333, 12.667]     7     22      24
             (12.667, 14]       NaN    NaN     NaN
       B     (9.996, 11.333]      2     11      10
             (11.333, 12.667]   NaN    NaN     NaN
             (12.667, 14]        14     33      40

如果需要，您可以 drop() strike，或者将 strikebins 替换为范围的平均值...

Answer 2

我猜 OP 想按分类变量分组，然后是一个按间隔分箱的数值变量。在这种情况下，您可以使用np.digitize()。

smallest = np.min(df['strike'])
largest = np.max(df['strike'])
num_edges = 3
# np.digitize(input_array, bin_edges)
ind = np.digitize(df['strike'], np.linspace(smallest, largest, num_edges))

那么ind应该是

array([1, 1, 2, 2, 2, 2, 3], dtype=int64)

对应分箱

 [10, 10, 12, 13, 12, 13, 14]

带 bin 边缘

array([ 10.,  12.,  14.]) # == np.linspace(smallest, largest, num_edges)

最后，按您想要的所有列进行分组，但要使用这个额外的 bin 列

df['binned_strike'] = ind
for grp in df.groupby(['symbol', 'serie', 'binned_strike']):
    print "group key"
    print grp[0]
    print "group content"
    print grp[1]
    print "============="

这应该打印出来

group key
('IP', 'A', 1)
group content
   last  price serie  strike symbol  type  binned_strike
0   1.0     11     A      10     IP  call              1
=============
group key
('IP', 'A', 2)
group content
   last  price serie  strike symbol  type  binned_strike
2   2.5     11     A      12     IP   put              2
4   4.5     11     A      12     IP  call              2
=============
group key
('IP', 'B', 1)
group content
   last  price serie  strike symbol type  binned_strike
1   2.0     11     B      10     IP  put              1
=============
group key
('IP', 'B', 2)
group content
   last  price serie  strike symbol type  binned_strike
3   3.0     11     B      13     IP  put              2
5   5.0     11     B      13     IP  put              2
=============
group key
('IP', 'B', 3)
group content
   last  price serie  strike symbol  type  binned_strike
6   6.0     11     B      14     IP  call              3
=============