hubs-predictor 0.1.2

Price predictions for 3DHubs

Price Prediction

Start Predicting

• Install the package:

  pip install hubs-predictor
  

• How to Use:

  from hubs_predictor import predict_batch, predict_from_csv
  
  # IP = 'localhost' if the container is running locally
  IP = '164.52.192.139' # Hosted in my private server for quick testing
  
  # Predict using CSV file:
  csv_path = '/PATH/TO/YOUR/CSV/FILE.csv'
  result = predict_from_csv(csv_path, ip=IP)
  
  # Predict as list batches
  # input format should be list of list [[], []]
  # eg: data = [['2015-01-09 23:15:03.308', 'cart__4', '27.917', '86.632', '11.022', '247384.01', ....., 'supplier__054'],
  #             ['2015-01-10 24:15:03.308', 'cart__5', '17.917', '76.632', '10.022', '147384.01', ....., 'supplier__055']]
  data = [[value1, value2, ...., value16],
          [value1, value2, ...., value16]]
  result = predict_batch(data, ip=IP)
  
  # NOTE: The input data both `csv_file` and `data` should be in the same feature sequence provided in the 'assignment-data.csv'
  # i.e: ['timestamp', 'cart', 'geometry/bounding_box/depth', 'geometry/bounding_box/width', ......., 'sourcing/supplier_country', 'sourcing/supplier']
  # which is 16 columns excluding `target/price`, if `target/price` is provided,  it will be ignored internally.
  

  Check Example script: test_package_example.py for more clarification.

  To build and setup local container please check Replicate / Reproduce Whole Process see below.

---

Replicate / Reproduce Whole Process

Training Process

• Pre-requirements:

  1. Install conda: link
  2. Install MLFlow:

     pip install mlflow
     

• Extract Data: assignment-data.zip to train/data/

  unzip assignment-data.zip -d train/data/
  

• Run Mlflow UI server to track the training experiments:

  cd train/
  mlflow server --backend-store-uri ./mlruns/ &
  
  # open 'http://localhost:5000/' to see the experiments
  # To stop mlflow server use `pkill -f mlflow`
  

• Train:

  # working dir: "ml-engineer-assignment-bendangnuksung/train/"
  
  # (OPTIONAL) modify "train/MLproject" file, update parameters such as:
  # 'datapath' -> path to your data CSV file
  # 'kfolds'   -> N kfolds you want
  # 'lr'       -> Set your own learning rate
  
  # Run training 
  mlflow run --experiment-name hubs_price_prediction .
  

Deployment Process

• Pre-requirements:

  1. Install Docker. Link
  2. Install Docker Compose:

     pip install docker-compose
     

• Build and Start Docker:

  # working dir: "ml-engineer-assignment-bendangnuksung/"
  
  # (OPTIONAL) Modify "docker-compose.yml" if:
  # 1. Wants to change PORT
  # 2. Change the volumes if model stored in different directory. (Default is: "./train/models" because models are stored there after training) 
  
  docker-compose up
  

• Building package (Optional):

  # modify setup.py accordingly
  python setup.py sdist
  python setup.py bdist_wheel
  twine check dist/*
  twine upload dist/*
  

Now the docker container is Up and Running, you can run the models locally. Just assign IP = 'localhost'

from hubs_predictor import predict_batch, predict_from_csv
IP = 'localhost'
csv_path = '/PATH/TO/YOUR/CSV/FILE.csv'
result = predict_from_csv(csv_path, ip=IP)
result = predict_batch(data, ip=IP)