Andikazidanef15 / Recommender-System-Project

Recommender System Project

Create menu's recommendation based on user's preference and user's customer segmentation by using Neural Collaborative Filtering (NCF) He et al., 2017 and K-Means Clustering . Due to sensitive information in the dataset, the data used in the recommender/cluster model is not stored in this data folder.

Pretrained NCF model available at lightning_logs/menus_df/checkpoints/epoch=49-step=504500.ckpt using NVIDIA T4 GPU on original data.

API Setup

Follow these steps to run the API code for generating sales and menus recommendation for particular user:

1. Insert the data (csv format) in the data folder
2. Install all required packages via terminal

pip install -r requirements.txt

3. Run main.py to start the flask API

python main.py

4. On seperate terminal, input User ID listed in the data and run this,

curl -X POST http://127.0.0.1:80/recms -H 'Content-Type: application/json' -d '{"user_id":"XXXXXXXX"}'

5. You can get the sales and menus recommendation in following format,

{'sales_id':['AAA','BBB',...], 'menu_id': ['CCC', 'DDD', ...]}

Training and Evaluate NCF Model

To train and evaluate NCF model run these steps:

1. Insert the data (csv format) in the data folder
2. Install all required packages via terminal

pip install -r requirements.txt

3. Run train.py and specify model parameters with arguments, for example if you want to train the model from beginning:

python train.py --path data/ --device cpu --num_factors 10 --num_hiddens '[10,10,10]' --num_neg 8 --epochs 50 --batch_size 16 --weight_decay 1e-5 --lr 0.01

If you want to use pretrained model, use this format

python train.py --device gpu --checkpoint lightning_logs/menus_df/checkpoints/epoch=49-step=504500.ckpt

All the checkpoints are stored in lightning_logs folder

4. Evaluate model by using this format

python evaluate.py --pretrained lightning_logs/menus_df/checkpoints/epoch=49-step=504500.ckpt

Notebook

For more detailed information about the analyis and code, you can run the Recommender_System.ipynb. Here you can find explanation for each steps in developing recommender system and cluster customers. Outputs are not shown because of sensitive information, you may want to include the data in data folder.

References

1. Neural Collaborative Filtering for Personalized Ranking
2. Deep Learning based Recommender System
3. Build a Movie Recommendation Engine backend API in 5 minutes (Part 2)