pytorch 多分类问题,计算百分比操作

Jenny ·

更新时间:2024-09-20

· 982 次阅读

二分类或分类问题，网络输出为二维矩阵：批次x几分类，最大的为当前分类，标签为one-hot型的二维矩阵：批次x几分类

计算百分比有numpy和pytorch两种实现方案实现，都是根据索引计算百分比，以下为具体二分类实现过程。

pytorch


out = torch.Tensor([[0,3],
     [2,3],
     [1,0],
     [3,4]])
cond = torch.Tensor([[1,0],
      [0,1],
      [1,0],
      [1,0]])
persent = torch.mean(torch.eq(torch.argmax(out, dim=1), torch.argmax(cond, dim=1)).double())
print(persent)

numpy


out = [[0, 3],
  [2, 3],
  [1, 0],
  [3, 4]]
cond = [[1, 0],
  [0, 1],
  [1, 0],
  [1, 0]] 
a = np.argmax(out,axis=1)
b = np.argmax(cond, axis=1)
persent = np.mean(np.equal(a, b) + 0)
# persent = np.mean(a==b + 0)
print(persent)

补充知识：python 多分类画auc曲线和macro-average ROC curve

最近帮一个人做了一个多分类画auc曲线的东西，不过最后那个人不要了，还被说了一顿，心里很是不爽，anyway，我写代码的还是要继续写代码的，所以我准备把我修改的代码分享开来，供大家研究学习。处理的数据大改是这种xlsx文件：


IMAGE y_real y_predict 0其他 1豹纹 2弥漫 3斑片 4黄斑
/mnt/AI/HM/izy20200531c5/299/train/0其他/IM005111 (Copy).jpg 0 0 1 8.31E-19 7.59E-13 4.47E-15 2.46E-14
/mnt/AI/HM/izy20200531c5/299/train/0其他/IM005201 (Copy).jpg 0 0 1 5.35E-17 4.38E-11 8.80E-13 3.85E-11
/mnt/AI/HM/izy20200531c5/299/train/0其他/IM004938 (4) (Copy).jpg 0 0 1 1.20E-16 3.17E-11 6.26E-12 1.02E-11
/mnt/AI/HM/izy20200531c5/299/train/0其他/IM004349 (3) (Copy).jpg 0 0 1 5.66E-14 1.87E-09 6.50E-09 3.29E-09
/mnt/AI/HM/izy20200531c5/299/train/0其他/IM004673 (5) (Copy).jpg 0 0 1 5.51E-17 9.30E-12 1.33E-13 2.54E-12
/mnt/AI/HM/izy20200531c5/299/train/0其他/IM004450 (5) (Copy).jpg 0 0 1 4.81E-17 3.75E-12 3.96E-13 6.17E-13

导入基础的pandas和keras处理函数

import pandas as pd

from keras.utils import to_categorical

导入数据

data=pd.read_excel('5分类新.xlsx')

data.head()

导入机器学习库


from sklearn.metrics import precision_recall_curve
import numpy as np
from matplotlib import pyplot
from sklearn.metrics import f1_score
from sklearn.metrics import roc_curve, auc

把ground truth提取出来

true_y=data[' y_real'].to_numpy()

true_y=to_categorical(true_y)

把每个类别的数据提取出来

PM_y=data[[' 0其他',' 1豹纹',' 2弥漫',' 3斑片',' 4黄斑']].to_numpy()

PM_y.shape

计算每个类别的fpr和tpr


n_classes=PM_y.shape[1]
fpr = dict()
tpr = dict()
roc_auc = dict()
for i in range(n_classes):
 fpr[i], tpr[i], _ = roc_curve(true_y[:, i], PM_y[:, i])
 roc_auc[i] = auc(fpr[i], tpr[i])

计算macro auc


from scipy import interp
# First aggregate all false positive rates
all_fpr = np.unique(np.concatenate([fpr[i] for i in range(n_classes)]))
# Then interpolate all ROC curves at this points
mean_tpr = np.zeros_like(all_fpr)
for i in range(n_classes):
 mean_tpr += interp(all_fpr, fpr[i], tpr[i])
# Finally average it and compute AUC
mean_tpr /= n_classes
fpr["macro"] = all_fpr
tpr["macro"] = mean_tpr
roc_auc["macro"] = auc(fpr["macro"], tpr["macro"])

画图


import matplotlib.pyplot as plt
from itertools import cycle
from matplotlib.ticker import FuncFormatter
lw = 2
# Plot all ROC curves
plt.figure()
labels=['Category 0','Category 1','Category 2','Category 3','Category 4']
plt.plot(fpr["macro"], tpr["macro"],
   label='macro-average ROC curve (area = {0:0.4f})'
    ''.format(roc_auc["macro"]),
   color='navy', linestyle=':', linewidth=4)
colors = cycle(['aqua', 'darkorange', 'cornflowerblue','blue','yellow'])
for i, color in zip(range(n_classes), colors):
 plt.plot(fpr[i], tpr[i], color=color, lw=lw,
    label=labels[i]+'(area = {0:0.4f})'.format(roc_auc[i]))
plt.plot([0, 1], [0, 1], 'k--', lw=lw)
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('1-Specificity (%)')
plt.ylabel('Sensitivity (%)')
plt.title('Some extension of Receiver operating characteristic to multi-class')
def to_percent(temp, position):
 return '%1.0f'%(100*temp)
plt.gca().yaxis.set_major_formatter(FuncFormatter(to_percent))
plt.gca().xaxis.set_major_formatter(FuncFormatter(to_percent))
plt.legend(loc="lower right")
plt.show()

展示