Review Dangerousness Classifier using ONNX

To determine whether an app is safe or not using binary classification based on app reviews. This model is BERT finetuned on a custom dataset to classify whether an app is "safe" or "not-safe" based on reviews given to it by users.

Steps to use this repo

Clone this repository and navigate to root folder

git clone https://github.com/lous-e/596E_Project1
cd 596E_Project1

Create virtual environment and install dependencies

Create a new virtual environment using any tool you prefer. We use venv for this example (on a Windows machine)

python -m venv venv
venv/Scripts/activate

Install the dependencies using the following command:

pip install -r requirements.txt

Download the model

Download the ONNX model from the following link: Model

Place the model in the root directory of this project.

Run the Flask-ML server

Run the following command to start the Flask-ML server (for Windows):

python -m server

Command line interface

To return the input string and model prediction in an output.json file

# text_to_classify is the input string, such as "this app is amazing"
python client.py --input_text [text_to_classify]

Steps to export ONNX model

Clone this repository found here

• Clone this repository and delete config.py
• Copy the files App_dangerrousness.py, bert_classifier.pth, config.py into the root directory of the folder.
• Configure Rescue Box Desktop to work with this using the instructions detailed below
• Uncomment the code at Lines 19, 57, 86 under "Original" and comment out the corresponding code under "ONNX".
• Run the server with the following command after following installation procedures

python -m server

The commented out code sets a breakpoint at Line 91 by adding the following line: import pdb; pdb.set_trace(). Send a request to the backend again using the same inputs from the RescueBox Desktop application. The breakpoint will be triggered in the backend. Run the following python code to export the ONNX model (also given in server.py)

torch.onnx.export(model, text, "review_classifier_model.onnx", export_params = True, opset_version = 16, do_constant_folding = True, input_names = ["input"], output_names = ["output"], dynamic_axes={'input' : {0 : 'batch_size'},'output' : {0 : 'batch_size'}})

The resulting ONNX model will be saved as review_classifier_model.onnx in the directory where server.py exists.

Download and run RescueBox Desktop from the following link: Rescue Box Desktop

Open the RescueBox Desktop application and register the model

Run the model

View the results

Attribution

This example uses a model that was originally created by a student in the Fall 2024 offering of 596E. The model has been modified to work with ONNX models. Their repo can be found here. Additionally, the code structure and documentation are based on this repo.

The base model is BERT, derived from the following paper:

@article{DBLP:journals/corr/abs-1810-04805,
  author       = {Jacob Devlin and
                  Ming{-}Wei Chang and
                  Kenton Lee and
                  Kristina Toutanova},
  title        = {{BERT:} Pre-training of Deep Bidirectional Transformers for Language
                  Understanding},
  journal      = {CoRR},
  volume       = {abs/1810.04805},
  year         = {2018},
  url          = {http://arxiv.org/abs/1810.04805},
  eprinttype    = {arXiv},
  eprint       = {1810.04805},
  timestamp    = {Tue, 30 Oct 2018 20:39:56 +0100},
  biburl       = {https://dblp.org/rec/journals/corr/abs-1810-04805.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}