# ML in Finance project
- Solution references:
- https://www.kaggle.com/code/tommy1028/lightgbm-starter-with-feature-engineering-idea/notebook
- https://www.kaggle.com/code/nyanpn/1st-place-public-2nd-place-solution
- Data snapshoot(of a certain stock id)
- book:
- trade: 

- Feature Engineering
- Book
- $wap1 = \frac{BidPrice1 * AskSize1 + AidPrice1 * BidSize1}{BidSize1 + AskSize1}$
- $wap2 = \frac{BidPrice2 * AskSize2 + AidPrice2 * BidSize2}{BidSize2 + AskSize2}$
- $logReturnWap1 = \Delta log(wap1)$
- $logReturnWap2 = \Delta log(wap2)$
- $logReturnAsk1 = \Delta log(AskPrice1)$
- $logReturnAsk2 = \Delta log(AskPrice2)$
- $logReturnBid1 = \Delta log(BidPrice1)$
- $logReturnBid2 = \Delta log(BidPrice2)$
- $wapBalance = |wap1 - wap2|$
- $priceSpread = \frac{(AskPrice1 - BidPrice1)}{\frac{AskPrice1 + BidPrice1}{2}}$
- $BidSpread = BidPrice1 - BidPrice2$
- $AidSpread = AskPrice1 - AskPrice2$
- $TotalVolume = AskSize1 + BidSize1 + AskSize2 + BidSize2$
- $VolumeImbalance = |AskSize1 + AskSize2 - BidSize1 - BidSize2|$
- Trade
- $logReturn1 = \Delta log(Price)$
- $numberOfDeals = len(data)$
- $tradeSizeSum = sum(Size)$
- $orderCountMean = mean(OrderCount)$
- The past N seconds feature of all of the above
- Target (this is weird in my opinion)
- $stockIdTargetEnc$: mean of target of other timeslots.