>Are you using Jupyter notebooks? Are you saving the model and how?
Nope, its a python script. I dont use Jupyter notebook as it eliminates (somewhat) the randomising of the training and test dataset. The most of the resources (time) is needed for splitting and fitting which is dependent on randomising and the 'freshness' of the run hence, I use a python script which runs every iteration cycle. Also, since the events happen daily, I have to update the training dataset almost daily hence having a Juypter notebook is not my choice of path forward. Yes, if (and when) user prompted results would be needed, I would go with Juypter notebooks as the target market is extremely finicky and will not wait for 10 minutes for the output.
>Are you doing this manually right now instead of a script?
I cannot cal it a manual process right now.. its copy-paste.. earlier I had to type the predictors.. which was fun at the start as I understood (somewhat) how the predictors affect the output but now, I just copy-paste from websources
>What decisions do the results in the spreadsheet inform or what actions do they trigger? How would I, as a consumer, use these results?
Guidance on how the event result can look like.. and the user is on their own on what bets they would like to place. e.g. If its a hockey match, what can the scoreline look like, how many fouls, etc. In NFL, what would each quarter look like, etc. I stay away from advising what bets to put. Its safer for me that way. e.g. if 1st quarer of a NFL event would be 6-1, I would present the output as is.. the user can place bets on what they would seem appropriate.
I downloaded and have curated historical sports data from various sources and figured out the 'simplest' parameters which work.. which at the moment is just the defaults. Model training is not much to be honest. I have a fairly fast processor so it takes about 10 minutes to run every iteration.
It has evolved from predicting just one dimension/feature but now, it predicts all of the variables it can train on (all training features)
Earlier, I used to manually input every predictor and then run the algorithim but now, I have figured out the copy-paste method which essentially cuts the pre-processing from around 2 hours to 10 minutes now. It gets a bit hectic giver the current nature of sports happening every day and increasing frequency of cycles. That part can be optimised via web scraping but not a priority at the moment but is a priority nevertheless.
Outputs are generated into a spreadsheet at the moment but would publish them onto an online platform later once I figure out the 'monetizing' and user-prompted running of the algorithim part.
WIP.. So its not 'failed' yet nor plan to..
Learnt coding because I had free time (COVID) and wanted to involve son into STEP aspects of coding. Then, I saw machine learning as a motivation for side hustle; learnt some of the better libraries out there and applied in the area of sports predictions. Saw incredible success and planned on pivoting to sports betting and now we have a steady consumption of our algorithim output.
So, we are currently 'sharing' our success with others but plan to monetize it later.
Figuring that part out.