独热编码处理文本属性
学习来源:click here
当数据中存在文本属性时,机器学习算法不便于处理文本属性,这时候需要把文本属性转换成数字。转换时,如果属性间存在顺序关系,例如:(冷,暖,热),可以直接使用整数编码;但当属性间没有顺序关系时,例如:(红, 绿, 蓝),则可使用独热编码。
独热编码:编码属性的值为1,其余属性的值为0
一、人工独热编码
from numpy import argmax
import numpy as np data = 'hello world' alphabet = 'abcdefghigklmnopqrstuvwxyz ' char_to_int = dict((c, i) for i, c in enumerate(alphabet)) int_to_char = dict((i, c) for i, c in enumerate(alphabet)) #整数编码 integer_encoded = [char_to_int[char] for char in data] print(integer_encoded) #独热编码 OneHot_Encoder = list() for i in integer_encoded: letter = [0 for _ in range(len(alphabet))] letter[i] = 1 OneHot_Encoder.append(letter) print(np.array(OneHot_Encoder)) #从独热编码恢复数据(argmax-返回最大值的索引) inverted = int_to_char[argmax(OneHot_Encoder[0])] print(inverted)
#output:
二、Scikit-Learn独热编码
from numpy import argmax from numpy import array from sklearn.preprocessing import LabelEncoder from sklearn.preprocessing import OneHotEncoder #整数编码 data = array(['cold', 'cold', 'warm', 'hot', 'warm', 'cold', 'hot', 'hot', 'warm', 'cold']) label_encoder = LabelEncoder() label_encoded = label_encoder.fit_transform(data) print(label_encoded) #独热编码 onehot_encoder = OneHotEncoder(categories='auto') onehot_encoded = onehot_encoder.fit_transform(label_encoded.reshape(-1, 1)) onehot = onehot_encoded.toarray() print(onehot) #恢复编码 state = label_encoder.inverse_transform([argmax(onehot[0, :])]) print(state)
#output: