Testing this model ?

[RohithYogi]

How do I test the model by loading the ones in checkpoints - like if I have a PE file - I can feed it to the model and get the result (true/false)?

[RohithYogi]

python -m src.bin.malshare - NOT WORKING
python train.py --benign_dir="../../raw/dll/" --malware_dir="../../raw/dasmalwerk/" - ERROR [Unpickling](https://github.com/jaketae/deep-malware-detection/issues/2)

python -m src.bin.dll
python extract_header.py --input_dir="../../raw/dll/" --output_dir="../../data/dll/"

python -m src.bin.dasmalwerk
python extract_header.py --input_dir="../../raw/dasmalwerk/" --output_dir="../../data/dasmalwerk/"

python train.py --benign_dir="/workspaces/deep-malware-detection/data/dll/" --malware_dir="/workspaces/deep-malware-detection/data/dasmalwerk/"

After I resolved the issues - I am now able to train the model - but that's it
If I give a new sample malware to benign then it should predict it right (a boolean) -

How can I test this ?

[jaketae]

Hi, I think what you're looking for is an evaluation/inference pipeline. I don't have a script specifically for that purpose, but you should be able to write one pretty quickly by importing these functions:

deep-malware-detection/src/deep_malware_detection/utils.py

Lines 76 to 100 in 8c45fc0

	def predict(model, data_loader, device, apply_sigmoid=False, to_numpy=True):
	model.eval()
	y_true = []
	y_pred = []
	for inputs, labels in tqdm(data_loader, leave=False):
	inputs = inputs.to(device)
	outputs = model(inputs)
	y_true.append(labels)
	y_pred.append(outputs)
	y_true = torch.cat(y_true).to(int)
	if apply_sigmoid:
	y_pred = sigmoid(torch.cat(y_pred))
	else:
	y_pred = (torch.cat(y_pred) > 0).to(int)
	if to_numpy:
	y_true = y_true.cpu().numpy()
	y_pred = y_pred.cpu().numpy()
	assert y_true.shape == y_pred.shape
	model.train()
	return y_true, y_pred
	def get_accuracy(model, data_loader, device):
	y_true, y_pred = predict(model, data_loader, device, to_numpy=False)
	return 100 * (y_true == y_pred).to(float).mean().item()

Specifically, you can load your model and create a data loader for your dataset to measure model accuracy.

To get a prediction for a single file, you can take a look at the predict function to see how you can get a boolean label given an input. Basically, if the output of the model is negative, it means a 0, otherwise, it is 1. If you prefer a probabilistic interpretation, you can also cast a sigmoid to the output.

Hope this helps!