mokarakaya/mlflow-azure-recommender

mlflow example project for azure recommender algorithms

mlflow-azure-recommender

mlflow example project for azure recommender algorithms

• This project runs mlflow experiments on SAR algorithm by using movielens100K data set.

• It saves the best model after hyperparameter tuning.

• Finally it serves the best model as a mlflow service.

• SAR algorithm is available at https://github.com/microsoft/recommenders

Start Mlflow server to visualize experiments

mlflow server     --backend-store-uri sqlite:///mlflow.db     --default-artifact-root ./artifacts     --host 0.0.0.0

Run training code

• train.py evaluates SAR algorithm by using different similarity metrics.
• Each experiment is saved to mlflow.
• Best model is saved as well.
• Remember to remove sar_best folder before running the script.
• Remember to make sure the dataset is available at ~/datasets/movielens100K/u.data before running the script.

python train.py

Sar model (similarity_type= jaccard ):
  RMSE: 3.7950736953280937
Sar model (similarity_type= lift ):
  RMSE: 4.046395609396664
model_input    user_id
0        1
model output:    user_id  item_id  prediction
0        1      346    0.006996
1        1      844    0.006996
2        1      769    0.006996
3        1      394    0.006996
4        1      345    0.006996
5        1      849    0.006996
6        1      812    0.006996
7        1      795    0.006996
8        1      912    0.006996
9        1     1682    0.006996

• Remember to add scipy to conda.yml since we need it in the server, but it is not automatically added to the auto-generated file.

• We can compare the experiment results on mlflow ui:

Serve the saved best model

mlflow models serve -m sar_best -p 5001

Test endpoint

curl http://127.0.0.1:5001/invocations -H 'Content-Type: application/json' -d '{
    "data": {"user_id": [1]}
}'