''' Test network trained by goldfish_orange_train '''
import numpy as np
import torch
from datareader import load_test_data
from neuralnet import NeuralNet
MODEL_PATH = 'neuralnet.pth'
neuralnet = NeuralNet()
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
neuralnet.to(device)
neuralnet.load_state_dict(torch.load(MODEL_PATH))
# Load and preprocess image
test_images, test_labels = load_test_data()
# Transform images to tensor and normalize them
mean = (255 - 0) / 2 # images contain values in range [0, 255]
test_images = (test_images.astype(np.float32) - mean) / mean
test_images = torch.tensor(test_images, device=device)
# Transform test labels to tensor
test_labels = torch.tensor(test_labels, device=device, dtype=torch.int64)
with torch.no_grad():
outputs = neuralnet(test_images)
_, predicted = torch.max(outputs.data, 1)
correct = (predicted == test_labels).sum().item()
