A machine learning based tool to predict the expected sale values of Cryptokitties.
- Collect initial data
- Connect to current kitties on sale!
- Create Update algorithm
- Collect trend data
- Collect eth price data
- Get predict data working
- Make a combine script (for train_dense, trend and eth price) - probably going to be a bit nasty
- Update trends has lots of errors - how to fix?
- Add columns that measures scarcity of cat? - NEED TO MEASURE ACTUAL SCARCITY, NOT JUST SCARCITY IN DATASET
- Make tool to see past cryptokittie sales that are similar
- Create model class which check if model saved avaliable or trains new model
- Connect email - send top 50 buys with important stats
- Log best 50 buy for each day
- Make everything into classes
- Make website to display everything?
- add logging
- figure out how to optimise model performance
- create main file to run everything
- create requirements.txt
- test on raspberry pi
- host website on raspberry pi!!!
How do you know if its actually a good prediction? or just a one off really shit one? Mewtation gems??
- Subactions will be called by a
main.pyscript. This will also organise which results to send via email.
Data
- Initial - collecting all of the initial data to be used, should only be run once
- Update - used to update the
train_dense.csvwith the latest data - Prediction - collecting the latest data for kitties on sale (~50,000) through steps of 500-1000
- Trends - collecting google trend data
- Eth_price - collecting ethereum price
- Combine - will combine the collected data (either for train or predict)
Model
- Train - use combined training data to train a model, output saved in
Saved_modelsandLabel_encoders - Predict - use combined predicting data to predict on data, output saved in
Predictions - Clean - run to delete excess saved models and label encoders (want to keep lightweight)
- Send - send an email with given contents
create a virtual env ect.... pip install -r requirements.txt
- create cool links Cryptokitties sale contract: https://etherscan.io/address/0xb1690c08e213a35ed9bab7b318de14420fb57d8c#tokentxnsErc721