AI Journey 2019: Combined Solution

Русская версия этого документа находится здесь.

This is a combined solution of AI Journey 2019 challenge. It consists of refactored code of top-20 solutions from the challenge. Its score is 69.

Knowledge base and models

Knowledge base of AI Journey 2019 contains data and models, which could be useful for AGI and applied NLP tasks:

• Unified State Exam solving;
• text summarization;
• text generation;
• style transfer;
• punctuation restoring;
• grammar error correction;
• domain-specific language modeling;
• discourse analysis;
• topic modeling;
• text classification.

To download the knowledge base please use:

python download_data.py

Directory models contains models and additional files for solvers of exam tasks.

To download models please use:

python download_models.py

Running a docker container

You can run a container with:

$ sudo docker run -w /workspace -v $(pwd):/workspace -p 8000:8000 -it alenush25/combined_solution_aij:latest python solution.py

It will run the container with HTTP-server on port 8000. It supports the following requests:

GET /ready

The return code will be 200 OK only if the solution is ready. Any other code means that the solution is not ready.

POST /take_exam

It is a request to begin the exam. Body of the request is a JSON object with an instance of exam test in JSON format (a sample JSON could be found in the folder test_data).

The solution should response to this request 200 OK and return a JSON-object with answers to the tasks.

Both the request and the response should have Content-Type: application/json. We recommend to use UTF-8 encoding.

We also publish a file metadata.json which was used for submission. Its content is below:

{
    "image": "alenush25/combined_solution_aij:latest",
    "entry_point": "python solution.py"
}

Where image — a field with docker-image name for the solution image, entry_point — a command which runs the solution. As a root directory the root of an archive with solution will be used.

A file eval_docker.py an example of upload and processing of an exam instance in JSON from the directory test_data. It then is sent to the solution

Solution Description

The task and solution description could be found here.