plot1.py

import argparse 
import math
import sys
import os 
parser=argparse.ArgumentParser()
parser.add_argument('-f',default='',type=str)
parser.add_argument('-fold',default=0,type=int)
parser.add_argument('-avg',default=1,type=int)
parser.add_argument('-test',default=1,type=int)
args=parser.parse_args()

files=[os.path.join(args.f,x) for x in os.listdir(args.f)]
test_acc={}
train_acc={}
train_cnt={}
test_cnt={}
layer_acc=[{},{},{},{}]

train_record=[]
test_record=[]
for i in range(1000):
    train_record.append([])
    test_record.append([])
max_layer=-1
flag=0

for x in files:
    f=open(x,'r')
    
    for l in f.readlines():
        if ('of epoch' in l) and ('layer0' in l):
            for k in range(100):
                if ('layer%d'%k) in l:
                    max_layer=max(max_layer,k+1)
            break
    f.close()
    
    

    f=open(x,'r')
    for l in f.readlines():
        if 'of epoch' not in l:
            continue
        words=l.strip().split(' ')
        epoch=int(words[words.index('epoch')+1][:-1])
        if epoch not in train_acc.keys():
            train_acc[epoch]=0
            test_acc[epoch]=0
            train_cnt[epoch]=0
            test_cnt[epoch]=0
            for k in range(max_layer):
                layer_acc[k][epoch]=0
        if args.avg:
            acc=float(words[words.index('avg_acc')+1])
        else:
            acc=float(words[words.index('acc')+1])
        if 'test' in l:
            test_cnt[epoch]+=1
            test_acc[epoch]+=acc
            test_record[epoch].append(acc)
            for k in range(max_layer):
                acc_k=float(words[words.index('layer%d'%k)+1])
                layer_acc[k][epoch]+=acc_k
        else:
            train_cnt[epoch]+=1
            train_acc[epoch]+=acc
            train_record[epoch].append(acc)
    flag=1


def var(l,avg):
    sqsum=0
    for x in l:
        sqsum+=(x-avg)**2
    return math.sqrt(sqsum/len(l)) 


max_acc=0
max_epoch=-1
for i,x in enumerate(sorted(train_acc.keys())):
    if not args.test:
        if train_cnt[x]<args.fold:
            continue
        acc=train_acc[x]/train_cnt[x]
        print('%d\t%.4f\t%.6f'%(i+1,acc,var(train_record[i+1],acc)))
    else:
        if test_cnt[x]<args.fold:
            continue
        acc=test_acc[x]/test_cnt[x]
        s='%d\t%.4f\t%.4f'%(i+1,acc,var(test_record[i+1],acc))
        for k in range(max_layer):
            s+='\t%.4f'%(layer_acc[k][x]/test_cnt[x])
        print(s)
    if max_acc<acc:
        max_acc=acc
        max_epoch=i+1
    print('max',max_acc)

print('max acc: %.4f, epoch %d'%(max_acc,max_epoch))