将字符串转换为唯一的数值 Pandas 或 Numpy

Question

我是 Python 和深度学习的初学者。对于大多数人来说，这可能很容易，但我该怎么做呢？

如何将以下对象转换为唯一的数值？

df['city'].unique()

array(['LIMA', 'VACAVILLE', 'CINCINNATI', 'GLASGOW', 'BOWLING GREEN',
   'LANCASTER', 'HOUSTON', 'SPRINGFIELD', 'RAPID CITY', 'FORT WORTH',
   'LAREDO', 'NEW YORK', 'CHARLESTON', 'PITTSBURGH',
   'WEST VALLEY CITY', 'CAYCE', 'HOT SPRINGS NATIO', 'CANTON',
   'FORT WAYNE', 'DU BOIS', 'DAYTON', 'MASON CITY', 'WASHINGTON',
   'LAKE OSWEGO', 'FAYETTEVILLE', 'SALT LAKE CITY', 'KNOXVILLE',
   'TURLOCK', 'MCALLEN', 'CENTERVILLE', 'ROCHESTER', 'OKLAHOMA CITY',
   'GAUTIER', 'DOYLESTOWN', 'ATLANTA', 'MEADVILLE', 'FORT MYERS',
   'ERIE', 'BEAUMONT', 'JACKSON', 'CLARKSVILLE', 'BETHLEHEM',
   'SAN ANTONIO', 'LAS VEGAS', 'ATHENS', 'SAN LUIS OBISPO', 'SEATTLE',
   'BRADENTON', 'TINLEY PARK', 'HUNTLEY', 'SYRACUSE', 'WHEELWRIGHT',
   'TOWSON', 'YONKERS', 'ARDEN HILLS', 'MARION', 'LIVONIA',
   'COLORADO SPRINGS', 'CURWENSVILLE', 'SAINT CHARLES', 'PETERSBURG',
   'SCOTTSDALE', 'SILVER SPRING', 'PORTLAND', 'BIRMINGHAM',
   'CEDARVILLE', 'CLERMONT', 'ASHEVILLE', 'SHREVEPORT', 'DRAPER',
   'WAVERLY', 'CANANDAIGUA', 'MOUNT PLEASANT', 'MARIETTA', 'MANKATO',
   'HARLINGEN', 'HATCH', 'MOBILE', 'POULSBO', 'GARDEN GROVE',
   'GIG HARBOR', 'OCONOMOWOC', 'MOUNT MORRIS', 'ORLANDO',
   'DODGE CITY', 'DILLSBURG', 'HUNTSVILLE', 'KANSAS CITY',
   'JACKSONVILLE', 'DULUTH', 'CITRUS HEIGHTS', 'ONEONTA', 'LOS LUNAS',
   'GIBSONIA', 'ROBINSON', 'VERNON HILLS', 'PHOENIX', 'DESTIN',
   'SHEPHERD', 'BROOKLYN', 'PLANO', 'WINTERS', 'JAMAICA', 'POWAY',
   'LEXINGTON', 'UPLAND', 'NEW ALBANY', 'GREENVILLE',
   'JEFFERSON CITY', 'ARLINGTON', 'BUFFALO', 'LOS ANGELES',
   'CHARLOTTE', 'WEST LAFAYETTE', 'GARY', 'COOPERSTOWN', 'GREAT BEND',
   'DAVISON', 'SMYRNA', 'MISSOURI CITY', 'MEMPHIS',
   'FORT WALTON BEACH', 'KISSIMMEE', 'BATAVIA', 'OLDSMAR', 'WYNNE',
   'ASHVILLE', 'FT BRAGG', 'TROY', 'SHAKER HTS', 'CLEVELAND HTS',
   'HAMBURG'], dtype=object)

我正在尝试使用这些数据训练模型。

Answer 1

IIUC 需要factorize:

df = pd.DataFrame({'city':list('abcddf')})
df['city1'] = pd.factorize(df['city'])[0]

或者转换成categoricals 得到codes：

df['city'] = pd.Categorical(df['city'])
df['city1'] = df['city'].cat.codes

---

print (df)

  city  city1
0    a      0
1    b      1
2    c      2
3    d      3
4    d      3
5    f      4

Answer 2

你也可以试试sklearn.preprocessing.LabelEncoder。如文档中所述，它对标签进行编码，其值介于 0 和 n_classes-1 之间。

from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()
df['city_num'] = le.fit_transform(df['city'])

print(df.head())
#             city  city_num
# 0           LIMA        72
# 1      VACAVILLE       122
# 2     CINCINNATI        21
# 3        GLASGOW        50
# 4  BOWLING GREEN        10
print(len(df.city.unique()))
# 132
print(len(set(df.city_num)))
# 132

然后您可以将数值列转换为指标列

from sklearn.preprocessing import OneHotEncoder

ohe = OneHotEncoder()
city_ind = ohe.fit_transform(df.city_num.values.reshape(-1, 1))

print(type(city_ind))
# <class 'scipy.sparse.csr.csr_matrix'>

print(city_ind.shape)
# (132, 132)

print(city_ind[0:2, ].toarray())
# [[ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.]
#  [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.
#    0.  0.  0.  0.  0.  0.]]