This repo contains code that runs on Linode's Shared CPU where a binary classifier machine learning model file is placed in Linode Object Storage, which is consumed by a Flask web application written in Python to do prediction, log the prediction status in Linode's MySQL server, and return results.
Download the Postman collection which has sample API endpoints and JSON request body
Request URL - http://194.195.115.181
NOTE: this is a http endpoint and doesn't have a domain name yet - runs on port 80
- LINODE_OBJECT_STORAGE_ACCESS_KEY - Linode Object Storage Access Key
- LINODE_OBJECT_STORAGE_SECRET_KEY - Linode Object Storage Secret Key
- MYSQL_HOST - MySQL host name
- MYSQL_USERNAME - MySQL username
- MYSQL_PASSWORD - MySQL user password
- LATEST_MODEL_VERSION - the latest version of the model (V1, V2, etc)
Get the prediction for fresh water.
mode(str, required)
Mode of operation, do prediction or just describe the model (predict or model_describe)
features_dict(dict. required if mode is predict)
Input features for detection
model_version(str, optional)
Version of the model you want to use (v1, v2 or latest)
Defaults to latest
get_probability(bool, optional)
If set to true, the request should return the probability of the classes
Defaults to false
get_feature_importance(bool, optional)
If set to true, returns the importance of each feature the model was trained on
Defaults to false
get_model_features(bool, optional)
If set to true, returns the feature columns the model was trained on
Defaults to false
skip_db_update(bool, optional)
If set to true, skips the prediction update to db and make the response time faster
Content-Type (required) string
Media type of the body sent to the API. (application/json)
NOTE: it's an public API endpoint so no authentication required. Never run public public endpoints in production
Sample request body for prediction:
{
"features_dict": {
"pH": "0.916054662638588",
"Iron": "0.61964963700558",
"Nitrate": "0.0",
"Chloride": "0.0",
"Lead": "0",
"Zinc": "0.9780321533559888",
"Turbidity": "0.2486518821452759",
"Fluoride": "0.6913182398790103",
"Copper": "0.96396750718677",
"Odor": "0.7721724045887509",
"Sulfate": "0.81345037627716",
"Chlorine": "0.966623674745241",
"Manganese": "0.011527500694864",
"Total Dissolved Solids": "0.36944624557778"
},
"model_version": "v1",
"mode": "predict",
"get_probability": true,
"get_feature_importance": true,
"get_model_features": true,
"skip_db_update": false
}
The response include the extracted features in JSON format.
{
"prediction": "safe to consume",
"predicted_class": 0,
"probability": {
"0": 0.5384615384615384,
"1": 0.46153846153846156
},
"feature_columns": [
"pH",
"Iron",
"Nitrate",
"Chloride",
"Lead",
"Zinc",
"Turbidity",
"Fluoride",
"Copper",
"Odor",
"Sulfate",
"Chlorine",
"Manganese",
"Total Dissolved Solids"
],
"feature_importance": [
0.1315288343051377,
0.05904120131425621,
0.06170211245003547,
0.1086330101463574,
0.005545970740731968,
0.031131607589077844,
0.09375709765287078,
0.05967779427995336,
0.08468771243873978,
0.0849028544693462,
0.03929514319469508,
0.060234075439392784,
0.1428010505057667,
0.03706153547363892
],
"response_time": 0.0127,
"log_source": "linode"
}