I am getting very slow inference rates by creating a test dataloader with batch size of 10k(Fashion MNIST)
Is there any better faster way to do the same
import numpy as np
import matplotlib.pyplot as plt
from torchvision import datasets,transforms
from torch import utils,nn,load,exp,autograd
count=0
class Model(nn.Module):
def init(self):
super().init()
self.l1 = nn.Linear(in_features=784,out_features=1000,bias=False)
self.l2= nn.Linear(in_features=1000,out_features=1000,bias=False)
self.l3 = nn.Linear(in_features=1000,out_features= 500,bias=False)
self.l4 = nn.Linear(in_features=500,out_features=200,bias=False)
self.l5 = nn.Linear(in_features=200,out_features=10,bias=False)
def forward(self,x):
x = x.view(x.shape[0],-1)
x = nn.functional.relu(self.l1(x))
x = nn.functional.relu(self.l2(x))
x = nn.functional.relu(self.l3(x))
x = nn.functional.relu(self.l4(x))
x = nn.functional.log_softmax(self.l5(x), dim=1)
return x
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.5,0.5,0.5), (0.5,0.5,0.5))])
test_dataset = datasets.FashionMNIST(‘~/.pytorch/F_MNIST_data/’, train = False, transform=transform, download = True)
test_dataloader = utils.data.DataLoader(test_dataset, shuffle=True, batch_size=10000)
def test(i):
global count
title_order = [‘T-shirt/top’,
‘Trouser’,
‘Pullover’,
‘Dress’,
‘Coat’,
‘Sandal’,
‘Shirt’,
‘Sneaker’,
‘Bag’,
‘Ankle Boot’]
dataiter = iter(test_dataloader)
images, labels = dataiter.next()
img = images[i]
lbl = labels[i]
with autograd.no_grad():
output = model.forward(img)
c_number=np.argmax(output)
if title_order[lbl]==title_order[c_number]:
count = count+1
model=Model()
model.load_state_dict(load(‘Adam_baseline.pth’))
model.eval()
images,labels = next(iter(test_dataloader))
for i in range(10000):
test(i)
if(i%100==0):
print(i,“images inferenced”)
acc=count/10000
print(“Predicted % for SGD baseline”,acc)
