# A silver solution(33rd)

I'm a newbie in Kaggle and it's actually a shock for me to reach top 2%...
My solution is based on Lingzhi's LGBM One Step Ahead with some modifications:
using 8 weeks of training data starting from May 31st, 2017
extract min, max, var, skew and kurt
calculate first order difference for the mean values
Categorical features(store, item): one hot encoding, count (I learn this trick from
[https://www.kaggle.com/xiaozhouwang/2nd-place-lightgbm-solution](https://www.kaggle.com/xiaozhouwang/2nd-place-lightgbm-solution)
)
averaged across different random seeds to stabilize model performance (also learned from Little Boat's solution)
I have posted my LGBM model on
[https://www.kaggle.com/deepdreamer/a-silver-solution-33rd](https://www.kaggle.com/deepdreamer/a-silver-solution-33rd)
. It scores 0.517 in private board.
Thanks Little Boat for sharing his solutions in Porto Seguro and Ceshine lee for sharing the framework. I really learn a lot from them.