Serving ML models

Run serving service

To start serving the trained ML model, you need the files with the saved ML model and transformations.

The following CLI command will create the API server with the saved model:

insolver_serving -model path_to_model -transforms path_to_transforms -transforms_module path_to_transforms_module -service flask

You can choose the server based on Flask or FastApi with the option service:

• -service flask

• -service fastapi

• -service django

The default endpoint is http://localhost:8000/predict (for django use http://localhost:8000/predict/), but you can change it with the parameters -ip and -port.

For example:

insolver_serving -model path_to_model -transforms path_to_transforms -transforms_module path_to_transforms_module  -service flask -ip 127.0.0.10 -port 5000

• Features of passing paths to models.
  When using Django, it’s important to pass the absolute path in full.
  When using Flask and FastsApi, you can only pass the name of the model,
  but on the condition that it lies in the root of the project.

Using serving service

After starting the ML serving service, you can use it via REST API.

Example of the request to http://127.0.0.10:5000/predict endpoint:

{
    "df": {
        "LicAge": {
            "162849": 57
        },
        "Gender": {
            "162849": 0
        },
        "MariStat": {
            "162849": 0
        },
        "DrivAge": {
            "162849": 31
        },
        "HasKmLimit": {
            "162849": 0
        },
        "BonusMalus": {
            "162849": 80
        },
        "RiskArea": {
            "162849": 6
        },
        "Age_m": {
            "162849": 31
        },
        "Age_f": {
            "162849": 18
        },
        "Age_m_2": {
            "162849": 961
        },
        "Age_f_2": {
            "162849": 324
        },
        "ClaimAmount": {
            "162849": 349.4447129909
        }
    }
}

You can also create a request from a random sample from InsolverDataFrame:

import pandas as pd

from insolver import InsolverDataFrame
from insolver.model_tools import download_dataset

download_dataset('US_Accidents')
df = pd.read_csv('datasets/US_Accidents_June20.csv', low_memory=False)
InsDataFrame = InsolverDataFrame(df)
request_data = InsDataFrame.sample_request(batch_size=1)

You can find a complete example with model training and serving here.