Project:

Model part of location intelligence project, including training, validation, predicting and feature importance.

Implentmented models:

Wide&Deep[2016] @ models/location_recommendation.py

The basice model is based on that structure, while the id embedding module is replace by our region model.

Requirements:

Must:

conda install pytorch torchvision cudatoolkit=9.0 -c pytorch
pip install pretrainedmodels scikit-learn tqdm opencv-python pandas pygeohash
pip install --upgrade scikit-image

Optional

pip install imgaug cvxpy cvxopt folium

Projects:

1. location recommendation: main_location_company.py

It build the relationship between location(item) and company(user). But each user only buy one item. It includes train, validation, predict. Reasoning part is move into main_location_company_model_based_reason.py.

nohup python3 -u main_location_company.py --model location_recommend_model_v6 --run_root result/location_recommend_model_v6_191113 --lr 0.01 --mode train --apps _191113.csv >lrm_5c.out 2>lrm_5c.err &
python3 main_location_company.py --model location_recommend_model_v6 --run_root result/location_recommend_model_v6_5city_191113 --lr 0.01 --mode predict_salesforce --ckpt model_loss_best.pt --apps _191113.csv

2. main_location_company_model_based_reason.py

It generate the feature importance of input layer by b.p. the error of last layer.

python3 main_location_company_model_based_reason.py --apps _191113.csv --pre_name sampled_ww_

3. get_embedding_feature.py

It generate the embedded vector for each location id after the model is trained.

4. main_location_intelligence_region.py

It build the relationship between location(item) and company(user).

Each user has several item. Because each company inside the circle of a building will be counted in.
Region model is used to replace id embedding so that it is easy for adding new buildings inside.
It includes train, validation, predict. For predict part, embedded feature of location/region need to be produced before ahead.

nohup python3 -u main_location_intelligence_region.py --run_root result/location_RSRBv5_191114 --model location_recommend_region_model_v5 --lr 0.01 --mode train --trainStep 1000 --batch-size 4 --n-epochs 160 >mlir_5.out 2>mlir_5.err &
python3 -u main_location_intelligence_region.py --run_root result/location_RSRBv5_191114 --model location_recommend_region_model_v5 --lr 0.01 --mode validate --trainStep 1000 --batch-size 4 --n-epochs 160
python3 main_location_intelligence_region.py --run_root result/location_RSRBv5_191114 --model location_recommend_region_model_v5 --lr 0.01 --mode predict --batch-size 1 --apps _191114.csv

5. get_embedding_feature_region.py