Add RNN example to perform DGA detection with LSTMs #240
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
|
shrit
approved these changes
Jan 9, 2025
cpp/lstm/dga_detection/lstm_test.cpp
Outdated
| @@ -0,0 +1,123 @@ | |||
| /** | |||
| * @file lstm_dga_detection_train.cpp | |||
There was a problem hiding this comment.
The file name is not identical to this one.
What is the point of this one as well ? given that we have the train file above ?
There was a problem hiding this comment.
Ack, I thought I removed it! Sorry about that. Fixed in 6169ab1.
| test_labels = requests.get( | ||
| "https://datasets.mlpack.org/mnist/t10k-labels-idx1-ubyte.gz") | ||
| progress_bar("test_labels.gz", test_labels) | ||
| ungzip("test_labels.gz", "test_labels.ubytes") | ||
|
|
|
This should wait for merge on those other 3 PRs. |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds an example that uses two RNNs to solve the DGA detection problem. In short it is a classification problem to determine whether or not a domain was generated by a domain generation algorithm.
The model uses two RNNs: one trained on benign domains, and one trained on malicious domains. For prediction, we compute the likelihood of a domain coming from each model, and then we take the more likely model as the class. This is the generalized likelihood ratio test or something very much like it. We used this strategy at Symantec and a more complicated version of it in a paper.
The network structure is simple: it uses
floatdata instead ofdoubledata and consists of 50 LSTM units followed by a 39-element linear layer and log-softmax.I plan to use this as a demonstration of how to deploy a predictive model inside of a Docker container.
When the model is trained and run, here is the output:
The size of the resulting models is small:
And the prediction program works like below, where I write a domain name and then the prediction is printed (or an error if the domain name was invalid):
For the code to run correctly, the following PRs must first be merged in mlpack: