-
Notifications
You must be signed in to change notification settings - Fork 2
/
trainer.py
129 lines (103 loc) · 5.21 KB
/
trainer.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
import mxnet as mx
from mxnet import gluon, autograd, nd
from mxnet.gluon import nn,utils
import mxnet.ndarray as F
import numpy as np
import os, sys
from tqdm import trange
import pickle
import random
from models import *
from utils import *
from data_loader import load_data
# set gpu count
def setting_ctx(GPU_COUNT):
if GPU_COUNT > 0 :
ctx = [mx.gpu(i) for i in range(GPU_COUNT)]
else :
ctx = [mx.cpu()]
return ctx
#Define Evaluation metric
def evaluate_accuracy(data, model, batch_size, ctx):
data_conv = cvt_data_axis(data)
acc = mx.metric.Accuracy()
accuracy_mat = []
for batch_idx in range(len(data) // (batch_size)):
input_img, input_qst, label = ndarray_conv(data_conv,batch_idx,batch_size)
input_img = input_img.as_in_context(ctx)
input_qst = input_qst.as_in_context(ctx)
label = label.as_in_context(ctx)
output = model(input_img,input_qst)
predictions = nd.argmax(output,axis=1)
acc.update(preds=predictions, labels=label)
accuracy_mat.append(acc.get()[1])
accuracy = sum(accuracy_mat) / len(accuracy_mat)
return accuracy
class Train(object):
def __init__(self, config):
##setting hyper-parameters
self.args = dict()
self.args['batch_size'] = config.batch_size
self.args['epoches'] = config.epoches
self.GPU_COUNT = config.GPU_COUNT
self.ctx = setting_ctx(self.GPU_COUNT)
self.show_status = config.show_status
self.build_model()
def build_model(self):
self.model = RN_Model(self.args)
#parameter initialozation
self.model.collect_params().initialize(ctx=self.ctx)
#set optimizer
self.trainer = gluon.Trainer(self.model.collect_params(),optimizer='adam',optimizer_params={'learning_rate':0.0001})
#define loss function
self.loss = gluon.loss.SoftmaxCrossEntropyLoss()
def train(self):
##load input data
rel_train, rel_test, norel_train, norel_test = load_data()
rel_loss = list()
norel_loss = list()
rel_acc = list()
noel_acc = list()
for epoch in trange(self.args['epoches']):
cumulative_rel_loss = 0.0
cumulative_norel_loss = 0.0
input_rel_train = rel_train.copy()
input_norel_train = norel_train.copy()
#shuffle data
random.shuffle(input_rel_train)
random.shuffle(input_norel_train)
rel = cvt_data_axis(input_rel_train)
norel = cvt_data_axis(input_norel_train)
#for batch_idx in tqdm(range(len(rel[0]) // (args['batch_size'] * 4))):
for batch_idx in range(len(rel[0]) // (self.args['batch_size']*self.GPU_COUNT)):
input_rel_img, input_rel_qst, rel_label = ndarray_conv(rel,batch_idx,self.args['batch_size']*self.GPU_COUNT)
#data split
input_rel_img = gluon.utils.split_and_load(input_rel_img,self.ctx)
input_rel_qst = gluon.utils.split_and_load(input_rel_qst,self.ctx)
rel_label = gluon.utils.split_and_load(rel_label,self.ctx)
coord_tensor = F.zeros((self.args['batch_size'] * self.GPU_COUNT, 25, 2))
coord_tensor = gluon.utils.split_and_load(coord_tensor,self.ctx)
with autograd.record():
rel_losses = [self.loss(self.model(X,Y),Z) for X, Y, Z in zip(input_rel_img,input_rel_qst,rel_label)]
for l in rel_losses:
l.backward()
self.trainer.step(self.args['batch_size']*self.GPU_COUNT)
for l in rel_losses:
cumulative_rel_loss += nd.sum(l).asscalar()
input_norel_img, input_norel_qst, norel_label = ndarray_conv(norel,batch_idx,self.args['batch_size']*self.GPU_COUNT)
#data split
input_norel_img = gluon.utils.split_and_load(input_norel_img,self.ctx)
input_norel_qst = gluon.utils.split_and_load(input_norel_qst,self.ctx)
norel_label = gluon.utils.split_and_load(norel_label,self.ctx)
with autograd.record():
norel_losses = [self.loss(self.model(X,Y),Z) for X, Y, Z in zip(input_norel_img,input_norel_qst,norel_label)]
for l in norel_losses:
l.backward()
self.trainer.step(self.args['batch_size']*self.GPU_COUNT)
for l in norel_losses:
cumulative_norel_loss += nd.sum(l).asscalar()
rel_accuracy = evaluate_accuracy(rel_test, self.model, self.args['batch_size'], mx.gpu(0))
norel_accuracy = evaluate_accuracy(norel_test, self.model, self.args['batch_size'], mx.gpu(0))
if(self.show_status):
if(epoch % 10 == 0):
print("Epoch {e}. rel_Loss: {rl} norel_Loss: {nrl} rel_ACC: {rl_acc} norel_ACC: {nrl_acc}".format(e=epoch+1, rl=cumulative_rel_loss/(len(rel[0]) // self.args['batch_size']), nrl=cumulative_norel_loss/ (len(rel[0]) // self.args['batch_size']), rl_acc=rel_accuracy,nrl_acc=norel_accuracy))