Inference API Specifications - Easy Predictive Analytics

The REST API can be used to run Easy Predictive Analytics predictions.

Inference API

Predictions can be run on prediction data using a previously created model. Create the endpoints (URLs), API keys, etc. required to run from the “API Creation Screen”. You can check the necessary information on the screen after the API is created.

URL

You can check the URL on the “API Creation Screen”.

The HTTP POST method is used.

Request

name	Description
x-api-key	API Key. One is assigned to each model.
content-type	Specify application/json; charset=utf-8

path parameter

Not used.

query parameter

Not used.

body

The format is indicated below.

{
	"inputs": [
		{
			"name": "pr",   # prediction data
			"type": "csv",
			"data": "6LfjeFlG/xxxxxx"
		},
		{
			"name": "ar",	# add analysis result
			"type": "boolean",
			"data": true
		},
		{
			"name": "ad",	# add prediction data
			"type": "boolean",
			"data": true
		},
		{
			"name": "nc",	# without classification results
			"type": "boolean",
			"data": true
		},
		{
			"name": "nr",	# without row numbers
			"type": "boolean",
			"data": true
		},
		{
			"name": "pc",
			"type": "string",
			"data": "column_name"
		}
	]
}

inputs parameter

name	type		Description
pr	csv(base64)	required	Value of the base64-encoded csv of the prediction file consisting of a row of the variable name of the data you wish to predict and one row of data
ar	boolean	required	Whether to calculate the reason for the prediction (output prediction_analysis.csv). Specify `true` if a prediction reason is required. Specify `false` to reduce the processing time without requiring a prediction reason.
ad	boolean	required	Whether to add prediction data to the output. Specify `true` to add it.
nc	boolean	required	Whether to output the classification results (in addition to the predicted probabilities) during binary and multiclass classification. Specify `true` if classification results are not required. `false` if classification results are required.
nr	boolean		Whether to output row numbers in the prediction results. Specify `true` if the row number is not required. Specify `false` if the row number is required.
pc	string		Name of the variable to be attached to the left end of the prediction result.Specify one of the variable names contained in the prediction file specified with pr.

Response

body

The format is indicated below.

{
    "version": str,
	"outputs": [
		{
			"name": "prediction_analysis.csv",
			"data": "6LfjeFlG/xxxxxx"
		},
		{
			"name": "prediction_valid.csv",
			"data": "6LfjeFlG/xxxxxx"
		}
	]
}

outputs parameter

name	type	Description
prediction_valid.csv	csv(base64)	Value of the base64-encoded csv of the prediction results
prediction_analysis.csv	csv(base64)	Value of the base64-encoded csv of the prediction reasons. Included only if ar is set to `true` in Request.

error message

code	error	message	reason
4001030101	NOT_FOUND_CLASSIFIER	Not found classifier in classifiers table	classifier_id is incorrect
4001030104	BAD_REQUEST	Invalid request	Schema error. Please make sure your input is in the correct format.
4031030201	ACCESS_DENIED	Invalid API key	The API key is incorrect
4031030201	ACCESS_DENIED	Invalid classifier id	classifier_id not found
4031030201	ACCESS_DENIED	Not allowed source ip	Unable to access source IP of connection
5031030101	SERVICE_UNAVAILABLE_ERROR	status is not running (status: stopped)	Server is not running

Execution Example

An example of executing the Inference API in Python is indicated below.

It is based on the assumption that the API has been created for a model created with the sample data “1_Premium Service Purchase.csv”. The prediction is calculated using the Inference API, using “2_Premium Service Purchase (for prediction).csv” as the prediction target.

import base64
import requests

prediction_data_file_path = '2_Premium Service Purchase (for prediction).csv'
api_url = 'YOUR_API_URL'
api_key = 'YOUR_API_KEY'

headers = {
    'x-api-key': api_key, 'content-type': 'application/json; charset=utf-8'}

with open(prediction_data_file_path, 'rb') as f:
    lines = f.readlines()

header = lines[0]
for line in lines[1:]:
    prediction_data = base64.b64encode(header + line).decode("ascii")
    body = {
        "inputs": [
            {
                "name": "pr",  # prediction data
                "type": "csv",
                "data": prediction_data
            },
            {
                "name": "ar",  # add analysis result
                "type": "boolean",
                "data": True
            },
            {
                "name": "ad",  # add prediction data
                "type": "boolean",
                "data": True
            },
            {
                "name": "nc",  # without classification results
                "type": "boolean",
                "data": True
            }
        ]
    }

    response = requests.post(api_url, headers=headers, json=body)
    response_json = response.json()
    if 'outputs' in response_json:
        for output in response_json['outputs']:
            print(output['name'])
            print(base64.b64decode(output['data']).decode())
    else:
        print(response_json)