yangge

## Runtime environment - **CPU**: 128 core - **memory**: 800GB - **GPU**: 8 * A100 80G You should install all dependencies with one command using: `pip install -r requirements.txt` ## !!! Firstly,Please `cd ./src` ## train loop #### 1. data update - Put the data for this competition in ../data then unzip it: https://www.kaggle.com/competitions/nfl-big-data-bowl-2021/data - Put the data for this competition in ../data then unzip it: https://www.kaggle.com/competitions/nfl-big-data-bowl-2026-prediction/data > The final data structure should be： > - `../data/nfl-big-data-bowl-2021` > - `../data/nfl-big-data-bowl-2026-prediction` #### 2. prepare data Process the data into training format and split it into 7 folds:` sh run_prepare.sh` #### 3. train models - run ` sh train.sh ` If you use our runtime environment configuration, it will likely take more than two days to run. ## infer loop Before making predictions,<br> please place the data to be predicted in the directory `../test_data/` <br> The data format should be consistent with the training data, including both `input_*.csv`and `output_*.csv` files. #### v1.After training is completed, you can make predictions using your trained ensemble model. Run this script directly: `python -u offline_final_sub.py`<br> The final saved prediction file is located in the .`./output_predict` folder<br> containing columns for both x and y coordinates, with each row corresponding to the respective predicted file. #### v2.Make predictions without the need to retrain the model. Run this script directly: `python -u offline_final_sub_v2.py`<br> The final saved prediction file is located in the .`./output_predict` folder<br> containing columns for both x and y coordinates, with each row corresponding to the respective predicted file. ## directory_structure.txt ``` . ./sub_model_save ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/fix_bug_pretrain_v4_fintune_v0_pretrain_2 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/fix_bug_pretrain_v4_fintune_v0_pretrain_4 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/fix_bug_pretrain_v4_fintune_v0_pretrain_6 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/fix_bug_pretrain_v4_fintune_v0_pretrain_3 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/fix_bug_pretrain_v4_fintune_v0_pretrain_5 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/fix_bug_pretrain_v4_fintune_v0_pretrain_0 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/fix_bug_pretrain_v4_fintune_v0_pretrain_1 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain/.ipynb_checkpoints ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090_2 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090_4 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090_6 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090_0 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090_1 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090_3 ./sub_model_save/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090/fix_bug_pretrain_v4_fintune_v0_pretrain_trains3090_5 ./sub_model_save/.ipynb_checkpoints ./sub_model_save/fix_bug_pretrain_decode_fintune_4090 ./sub_model_save/fix_bug_pretrain_decode_fintune_4090/fix_bug_pretrain_decode_fintune_4090_0 ./sub_model_save/fix_bug_pretrain_decode_fintune_4090/fix_bug_pretrain_decode_fintune_4090_3 ./sub_model_save/fix_bug_pretrain_decode_fintune_4090/fix_bug_pretrain_decode_fintune_4090_4 ./sub_model_save/fix_bug_pretrain_decode_fintune_4090/fix_bug_pretrain_decode_fintune_4090_5 ./sub_model_save/fix_bug_pretrain_decode_fintune_4090/fix_bug_pretrain_decode_fintune_4090_1 ./sub_model_save/fix_bug_pretrain_decode_fintune_4090/fix_bug_pretrain_decode_fintune_4090_2 ./sub_model_save/fix_bug_pretrain_decode_fintune_4090/fix_bug_pretrain_decode_fintune_4090_6 ./sub_model_save/fix_bug_pretrain_decode_fintune ./sub_model_save/fix_bug_pretrain_decode_fintune/fix_bug_pretrain_decode_fintune_5 ./sub_model_save/fix_bug_pretrain_decode_fintune/fix_bug_pretrain_decode_fintune_6 ./sub_model_save/fix_bug_pretrain_decode_fintune/fix_bug_pretrain_decode_fintune_4 ./sub_model_save/fix_bug_pretrain_decode_fintune/fix_bug_pretrain_decode_fintune_2 ./sub_model_save/fix_bug_pretrain_decode_fintune/fix_bug_pretrain_decode_fintune_3 ./sub_model_save/fix_bug_pretrain_decode_fintune/fix_bug_pretrain_decode_fintune_1 ./sub_model_save/fix_bug_pretrain_decode_fintune/fix_bug_pretrain_decode_fintune_0 ./test_data ./test_data/.ipynb_checkpoints ./logs ./logs/.ipynb_checkpoints ./output_predict ./output_predict/.ipynb_checkpoints ./model_save ./src ./src/.ipynb_checkpoints ./.ipynb_checkpoints ./data ./data/.ipynb_checkpoints ./output_data ./output_data/.ipynb_checkpoints ```