對于機(jī)器學(xué)習(xí)訓(xùn)練的模型而言,模型的準(zhǔn)確率,召回率和F1值是評價一個模型是否優(yōu)秀的參考。那么在pytorch中怎么計算準(zhǔn)確率,召回率和F1值呢?來看看小編是怎么做的。
看代碼吧~
predict = output.argmax(dim = 1)
confusion_matrix =torch.zeros(2,2)
for t, p in zip(predict.view(-1), target.view(-1)):
confusion_matrix[t.long(), p.long()] += 1
a_p =(confusion_matrix.diag() / confusion_matrix.sum(1))[0]
b_p = (confusion_matrix.diag() / confusion_matrix.sum(1))[1]
a_r =(confusion_matrix.diag() / confusion_matrix.sum(0))[0]
b_r = (confusion_matrix.diag() / confusion_matrix.sum(0))[1]
補(bǔ)充:pytorch 查全率 recall 查準(zhǔn)率 precision F1調(diào)和平均 準(zhǔn)確率 accuracy
看代碼吧~
def eval():
net.eval()
test_loss = 0
correct = 0
total = 0
classnum = 9
target_num = torch.zeros((1,classnum))
predict_num = torch.zeros((1,classnum))
acc_num = torch.zeros((1,classnum))
for batch_idx, (inputs, targets) in enumerate(testloader):
if use_cuda:
inputs, targets = inputs.cuda(), targets.cuda()
inputs, targets = Variable(inputs, volatile=True), Variable(targets)
outputs = net(inputs)
loss = criterion(outputs, targets)
# loss is variable , if add it(+=loss) directly, there will be a bigger ang bigger graph.
test_loss += loss.data[0]
_, predicted = torch.max(outputs.data, 1)
total += targets.size(0)
correct += predicted.eq(targets.data).cpu().sum()
pre_mask = torch.zeros(outputs.size()).scatter_(1, predicted.cpu().view(-1, 1), 1.)
predict_num += pre_mask.sum(0)
tar_mask = torch.zeros(outputs.size()).scatter_(1, targets.data.cpu().view(-1, 1), 1.)
target_num += tar_mask.sum(0)
acc_mask = pre_mask*tar_mask
acc_num += acc_mask.sum(0)
recall = acc_num/target_num
precision = acc_num/predict_num
F1 = 2*recall*precision/(recall+precision)
accuracy = acc_num.sum(1)/target_num.sum(1)
#精度調(diào)整
recall = (recall.numpy()[0]*100).round(3)
precision = (precision.numpy()[0]*100).round(3)
F1 = (F1.numpy()[0]*100).round(3)
accuracy = (accuracy.numpy()[0]*100).round(3)
# 打印格式方便復(fù)制
print('recall'," ".join('%s' % id for id in recall))
print('precision'," ".join('%s' % id for id in precision))
print('F1'," ".join('%s' % id for id in F1))
print('accuracy',accuracy)
補(bǔ)充:Python scikit-learn,分類模型的評估,精確率和召回率,classification_report
分類模型的評估標(biāo)準(zhǔn)一般最常見使用的是準(zhǔn)確率(estimator.score()),即預(yù)測結(jié)果正確的百分比。
混淆矩陣:
準(zhǔn)確率是相對所有分類結(jié)果;精確率、召回率、F1-score是相對于某一個分類的預(yù)測評估標(biāo)準(zhǔn)。
精確率(Precision):預(yù)測結(jié)果為正例樣本中真實為正例的比例(查的準(zhǔn))( )
召回率(Recall):真實為正例的樣本中預(yù)測結(jié)果為正例的比例(查的全)( )
分類的其他評估標(biāo)準(zhǔn):F1-score,反映了模型的穩(wěn)健型
demo.py(分類評估,精確率、召回率、F1-score,classification_report):
from sklearn.datasets import fetch_20newsgroups
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.metrics import classification_report
# 加載數(shù)據(jù)集 從scikit-learn官網(wǎng)下載新聞數(shù)據(jù)集(共20個類別)
news = fetch_20newsgroups(subset='all') # all表示下載訓(xùn)練集和測試集
# 進(jìn)行數(shù)據(jù)分割 (劃分訓(xùn)練集和測試集)
x_train, x_test, y_train, y_test = train_test_split(news.data, news.target, test_size=0.25)
# 對數(shù)據(jù)集進(jìn)行特征抽取 (進(jìn)行特征提取,將新聞文檔轉(zhuǎn)化成特征詞重要性的數(shù)字矩陣)
tf = TfidfVectorizer() # tf-idf表示特征詞的重要性
# 以訓(xùn)練集數(shù)據(jù)統(tǒng)計特征詞的重要性 (從訓(xùn)練集數(shù)據(jù)中提取特征詞)
x_train = tf.fit_transform(x_train)
print(tf.get_feature_names()) # ["condensed", "condescend", ...]
x_test = tf.transform(x_test) # 不需要重新fit()數(shù)據(jù),直接按照訓(xùn)練集提取的特征詞進(jìn)行重要性統(tǒng)計。
# 進(jìn)行樸素貝葉斯算法的預(yù)測
mlt = MultinomialNB(alpha=1.0) # alpha表示拉普拉斯平滑系數(shù),默認(rèn)1
print(x_train.toarray()) # toarray() 將稀疏矩陣以稠密矩陣的形式顯示。
'''
[[ 0. 0. 0. ..., 0.04234873 0. 0. ]
[ 0. 0. 0. ..., 0. 0. 0. ]
...,
[ 0. 0.03934786 0. ..., 0. 0. 0. ]
'''
mlt.fit(x_train, y_train) # 填充訓(xùn)練集數(shù)據(jù)
# 預(yù)測類別
y_predict = mlt.predict(x_test)
print("預(yù)測的文章類別為:", y_predict) # [4 18 8 ..., 15 15 4]
# 準(zhǔn)確率
print("準(zhǔn)確率為:", mlt.score(x_test, y_test)) # 0.853565365025
print("每個類別的精確率和召回率:", classification_report(y_test, y_predict, target_names=news.target_names))
'''
precision recall f1-score support
alt.atheism 0.86 0.66 0.75 207
comp.graphics 0.85 0.75 0.80 238
sport.baseball 0.96 0.94 0.95 253
...,
'''
召回率的意義(應(yīng)用場景):產(chǎn)品的不合格率(不想漏掉任何一個不合格的產(chǎn)品,查全);癌癥預(yù)測(不想漏掉任何一個癌癥患者)
以上就是在pytorch中計算準(zhǔn)確率,召回率和F1值的操作,希望能給大家一個參考,也希望大家多多支持W3Cschool。