Note that because of nature of project it could be very dependant to iterarions and no plan is strait forward for this type of projects.
- Add Dockerization for code Executor
- Explainablity
- Logging
- Getting formatted evaluation and code execution result without execution
- adding code generation to graph
Welcome to the Kaggle Problem Solver, the Swiss Army knife of machine learning challenges! This isn't just any old problem solver β it's your AI-powered companion in the wild world of Kaggle competitions. Using a "plan and execute" strategy that would make any project manager jealous, our system tackles ML problems with the finesse of a seasoned data scientist and the tireless energy of a thousand interns. code generation agent is inspired from langgraph agent link
- The Mastermind (KaggleProblemPlanner): Plans your path to Kaggle glory!
- The Perfectionist (KaggleTaskEnhancer): Turns good tasks into great ones!
- The Code Wizard (CodeGenerationAgent): Conjures code like magic! β¨
- The Strategist (KaggleProblemRePlanner): Adapts faster than a chameleon in a rainbow!
- The Executor (KaggleCodeExecutor): Runs code faster than you can say "machine learning"!
- Plan β 2. Enhance β 3. Code β 4. Execute β 5. Re-Plan β Repeat!
It's like a never-ending dance party, but with more algorithms and less awkward small talk.
Behold, the piΓ¨ce de rΓ©sistance of our project β the Agent Graph! π
This isn't just any graph β it's a visual symphony of our agents working in harmony. Watch as data flows through our system like a well-choreographed ballet of bits and bytes!
-
Clone this repo faster than you can say "git":
git clone https://github.com/msnp1381/kaggle-problem-solver.git
-
Let Poetry work its magic:
poetry install
-
Whisper your secrets to the
.env
file:HTTP_PROXY_URL=super_secret_proxy LANGFUSE_PUBLIC_KEY=shh_its_a_secret LANGFUSE_SECRET_KEY=dont_tell_anyone LANGFUSE_HOST=probably_not_localhost OPENAI_API_KEY=get_it_from_other_github_repos
-
Unleash the AI:
poetry run python agent.py
Customize your config like you're picking toppings for a pizza:
config = {
"callbacks": [langfuse_handler],
}
- Create new agents like you're assembling an AI Avengers team.
- Integrate them into
agent.py
β it's like introducing your new friends to your old crew. - Update
KaggleProblemSolver
to include your new agent in the coolest workflow in town.
Why stop at one graph when you can have a whole gallery? Here are some ideas to make your project visually stunning:
- The Task Lifecycle Rollercoaster: Show a task's journey from boring idea to ML masterpiece!
- The Metrics Dashboard of Dreams: Real-time stats that would make any data viz enthusiast weep with joy.
- Code Generation: The Movie: Animate your code generation process. It's like watching a blockbuster, but with more semicolons.
- The Interactive Notebook of Wonders: Let users play with your notebooks. It's like Jupyter, but cooler.
- The Dependency Web: Show how your tasks are connected. It's like LinkedIn, but for algorithms.
Got ideas? We want them! Check out CONTRIBUTING.md
for how to join our merry band of AI enthusiasts. Remember, in this repo, there are no bad ideas, only "learning opportunities"!
This project is licensed under the MIT License - see the LICENSE
file for details. In other words, go wild, but don't forget to give us a high-five if you use it!
Remember, in the world of Kaggle Problem Solving, the only limit is your imagination (and maybe your GPU's memory)! Happy coding, and may the algorithms be ever in your favor! π