Current Competition:

• Predict futures sales
• Timeline: 2 months（01/20/2021）
• Current progress(leaderboard): 0.986 rmse.
• Next step: read

Notebooks

Summary:

• 2nd place summary - Predict future sales
• 1st place solution gratitudes

EDA:

• Time series Basics : Exploring traditional TS
• Model stacking, feature engineering and EDA
• 1st place solution part 1

Features ideas and Ensemble:

• Future Sales 3
• Predict Future Sales Top 11 Solution
• Feature Engineer and Model Ensemble
• Feature engineering, xgboost
• Dance with Ensemble” Sharing Thread

DONE

• Regenerate my baseline from scratch(hold-out validation)
  1. W/features: shop_id, item_id, date_num_block.
  2. Lag … Lagged item_cnt_month for month/shop/item
  3. Date … Month
  4. Price … Average and variance of price for month/shop/item

Note: Switching to google colab from kaggle notebook for more consistent performance.

Result:

• Current baseline model: 0.986(public score)
  • model: xgb
  • completed: 12/13/2020

Next:

1. Re-implement the “Future Sale 3” notebook
   • Future Sales 3
   • ETA: by 12/20/2020

---