'''

实现对训练好的模型每个不同角度进行准确率比较。

'''

import torch.nn as nn

import torch.optim as optim

import torch.nn.functional as F

from torch.autograd import Variable

import numpy

import torch.utils.data as Data

import torch

import sys

import os

from ProcessProgram.NeurNetWorkProcess.ProceRawDataToTensor import *

Work_Path = "F:\\PlantarPressurePredictExperiment"

os.chdir(Work_Path)

#region ModelImport

#LSTM

# from ProcessProgram.Pytorch.LSTM import LSTM

# from ProcessProgram.Pytorch.LSTM import sequenceLen

# from ProcessProgram.Pytorch.LSTM import getmodel,BATCH_SIZE

# CNN

from ProcessProgram.Pytorch.CNN import CNN

from ProcessProgram.Pytorch.CNN import getmodel,BATCH_SIZE

#endregion

global x_test,y_test,test_loader

#region 文件导入

def importData():

SaveTestDataFilePath = "Pytorch\\data\\angle\\TestData.csv"

SaveTestLabelFilePath = "Pytorch\\data\\angle\\TestLabel.csv"

global x_test,y_test

with open(SaveTestDataFilePath, encoding='utf-8') as path_x_test_m, \

open(SaveTestLabelFilePath, encoding='utf-8')as path_y_test_m:

x_test = np.loadtxt(path_x_test_m, dtype=float,delimiter=",")

y_test = np.loadtxt(path_y_test_m,dtype = int, delimiter=",")

#endregion

#region 数据转换

def TrainFormData():

global x_test,y_test,test_loader

#LSTM RESHAPE

# x_test = x_test.reshape(len(x_test),sequenceLen,1260)

#CNN RESHAPE

x_test = x_test.reshape(len(x_test),1,60,21)

x_test = torch.from_numpy(x_test)

y_test = torch.from_numpy(y_test)

x_test = torch.tensor(x_test,dtype= torch.float32)

y_test = torch.tensor(y_test,dtype=torch.long)

test_dataset = Data.TensorDataset(x_test,y_test)

test_loader = Data.dataloader.DataLoader(dataset= test_dataset,batch_size = BATCH_SIZE ,shuffle= True)

#endregion

def TestNetWork(model):

correct = 0

correctCount = [0,0,0,0,0] #0，30，60，90 正确数量

TotalCount = [0,0,0,0,0] #总数

ErrRect = np.zeros((5,5))

for step, (data, target) in enumerate(test_loader):

data, target = Variable(data),Variable(target)

output = model(data)

# sum up batch loss

# get the index of the max log-probability

pred = torch.max(output.data, 1)[1]

correct += int(pred.eq(target.data.view_as(pred)).cpu().sum())

for ansIndex in range(len(pred)):

angleIndex = target[ansIndex] #真值

predictIndex = pred[ansIndex] #预测值

if(predictIndex == angleIndex):

correctCount[angleIndex] += 1

TotalCount[angleIndex] += 1

else:

TotalCount[angleIndex] += 1

ErrRect[angleIndex][pred[ansIndex]] +=1

correctRate = 100. * correct / len(test_loader.dataset)

print('Accuracy: {}/{} ({:.0f}%)\n'.format(correct, len(test_loader.dataset),correctRate))

print("Count：0°:{}/{} 30°:{}/{} 60°:{}/{} 90°:{}/{}".format(correctCount[0],TotalCount[0],

correctCount[1],TotalCount[1],

correctCount[2],TotalCount[2],

correctCount[3],TotalCount[3],))

print("angelAccuracy：0°:{:.2f}% 30°:{:.2f}% 60°:{:.2f}% 90°:{:.2f}%".format(100. * correctCount[0]/TotalCount[0],

100. * correctCount[1]/TotalCount[1],

100. * correctCount[2]/TotalCount[2],

100. * correctCount[3]/TotalCount[3]))

print("错误矩阵:-------------\n",ErrRect)

return ("{:.0f}%".format(100. * correct / len(test_loader.dataset)))

if __name__ == '__main__':

importData()

TrainFormData()

lstm = getmodel("cnn3_31_0_correct88%.pkl") #放入模型

TestNetWork(lstm)

pass