README.md

multiclassify-eval

A simple python module to calculate precision, recall, accuracy and f-measure for multi-classify evaluation.

一个基于Python的简单的多分类问题评估模块，包含查准率(精确率)、查全率(召回率)、精确度，F1度量。

Introdution

You can init it by:

你可以通过以下方式初始化：

from evaluation import 
evals = Evaluations(pred,gt,CLASSES)

Which predand gt can be a 1-dimension numpy.ndarray or list range from $[0, N-1]$, $N$ is the length of CLASSES . And CLASSES is a 1-dimension list contains class labels and len(CLASSES)=N must be hold.

pred和 gt 应该是范围在$[0, N-1]$ 之间的一维numpy.ndarray或list，即numpy或列表，其中$N$是CLASSES的长度。CLASSES 是一个包含标签的一维list。

After that you can call evals.average or getattr(evals,CLASSES[*]) with .precision(), .recall(), .accuracy() or .f1_score() to access average or any single class evaluation.

你可以通过下述方法来得到平均的或某个类别的分类情况。

precision_ = evals.average.precision()
recall_ = evals.average.recall()

class_ = getattr(evals,CLASSES[*])
accuracy_ = class_.accuracy()
accuracy_ = class_.f1_score()

And you can call print(evals) to see all classes.

你也可以通过print()在控制台中显示所有类别的分类情况。

---

Demo

A simple demo to explain how to use this module in pytorch if you are dealing with a multi classification problem.

如果你在pytorch中实现一个多分类问题，这是一个简单的实例指明如何使用该模块。

from evaluation import *

#Your classes from 'A' to 'Z'
CLASSES = [chr(65 + i_) for i_ in range(26)]

...

#list to put pred & groundtruth labels
pred_list = []
gt_list = []

#some loop to train your data
for i_batch, batch_ in enumerate(dataloader):
  
  ...
  
  #get gt labels
  label = sample_batched['label'].squeeze(-1).to(device)
  gt_list.append(label.detach().cpu().numpy().tolist())
  
  ...
  
  #get pred labels
  pred_ = net(data)
  pred_list.append(pred_.argmax(dim=1).detach().cpu().numpy().tolist())

#transform list to np.ndarray
pred_np = np.array(pred_list)
gt_np = np.array(gt_list)

evals = Evaluations(pred_np,gt_np,CLASSES)

print(evals.average.precision())
print(evals.A.recall())
...

#if you are using tensorboardX in pytorch, you can call this to record data
evals.writelog(writer_test,path='validation',global_step=i_epoch)

---

TensorboardX

if you are using tensorboardX in pytorch, you can call evals.writelog() to record data.

如果你在使用tensorboardX，你也可以用通过 evals.writelog() to来记录数据。

def writelog(self,writer,key='average',path='',global_step = None):

writer should be a summarywriter pre-defined. key can be any item in CLASSES ,'ALL' or 'averager'

writer 应该被预定义好，key可以是CLASSES中的任意元素，'ALL' 或者 'averager' 。