First daap for implementation of Private Model, Public Input data science competitions
Organizer
- Puts training data publicly offchain (GDrive, IPFS, ...)
- Initiates the competition by deploying the contract
- Prepares test data offchain (need to be defined before the competition begins, but not viewable, otherwise participants would overfit it)
- Calculates merkle root of test data
- Commits testing data with that merkle root
Client can use this Jupyter notebook to deploy/interact with the contracts
This phase last 1 week from the moment organizer commited test data
Competitors
- Use training data to train model
- Calculate hash of the model
- Commit model with that hash
Client can use this Jupyter notebook to calculate hash, commit model to the contract. Interacting with contract by Jupyter notebook makes the it seamlessly integrated to machine learning workflow.
Starts when phase 1 finished
The Jupyter notebooks mentioned above also support these functions invocation
Competitors
- Reveal model
Organizer
- Reveal testing data
Now all submission models and testing data are available onchain Anyone can call evaluation function calculate all competitors evaluation metrics and rank them.