logistics多分类

multiclassification

#DATASET: https://archive.ics.uci.edu/ml/datasets/Glass+Identification
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import sklearn 
import sklearn.preprocessing as pre

df=pd.read_csv('data\glassi\glass.data')
X,y=df.iloc[:,1:-1],df.iloc[:,-1]
X,y=np.array(X),np.array(y)

for idx,class_name in enumerate(sorted(list(set(y)))):
    y[y==class_name]=idx
    
from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.15,random_state=66)

f_mean, f_std = np.mean(X_train, axis=0), np.std(X_train, axis=0)
X_train = (X_train - f_mean) / f_std
X_test = (X_test - f_mean) / f_std

#add a constant parameter
X_train = np.concatenate((np.ones((X_train.shape[0], 1)), X_train), axis=1)
X_test = np.concatenate((np.ones((X_test.shape[0], 1)), X_test), axis=1)

#gradient descent function

def get_classifier(X_train,y_train,num_epoch=10000,alpha=0.01):
    theta=np.zeros(X_train.shape[1])
    for epoch in range(num_epoch):
        logist=np.dot(X_train,theta)
        h=1/(1+np.exp(-logist)) #hypothesis function
        cross_entropy_loss=(-y_train*np.log(h)-(1-y_train)*np.log(1-h)).mean()
        gradient=np.dot((h-y_train),X_train)/y_train.size
        theta-=alpha*gradient #update
    return theta

def multi_classifier(X_train,y_train):
    num_class=np.unique(y_train)
    parameter=np.zeros((len(num_class),X_train.shape[1])) #each has an array of parameters
    for i in num_class:       
        label_t=np.zeros_like(y_train) #use label_t to label the target class!!!
        num_class=np.unique(y_train)
        label_t[y_train==num_class[i]]=1 #important, 
        parameter[i,:]=get_classifier(X_train,label_t) #each array stands for one class's parameter
    return parameter

params = multi_classifier(X_train, y_train)

def pred(parameter,X_test,y_test):
    f_size=X_test.shape
    l_size=y_test.shape
    assert (f_size[0]==l_size[0])
    logist=np.dot(X_test,np.transpose(parameter)).squeeze()
    prob=1/(1+np.exp(-logist))
    pred=np.argmax(prob,axis=1)
    accuracy = np.sum(pred == y_test) / l_size[0] * 100   
    return prob, pred, accuracy

_, preds, accu = pred(params, X_test, y_test)
print("Prediction: {}\n".format(preds))
print("Accuracy: {:.3f}%".format(accu))

Prediction: [0 1 0 4 1 5 1 0 0 1 0 1 0 0 5 1 1 1 1 0 5 4 0 1 5 0 0 1 1 0 3 1 0]

Accuracy: 66.667%