Getting Started With ML Prediction Engine

Step 1: Install and Run the Engine

Both install paths start the same way: copy config.sample.php to config.php and set apiKey and adminPassword to long random strings.

Docker is the fastest path. Set mlServiceUrl to http://ml:8750 in the config, then:

cp config.sample.php config.php
# edit config.php: set apiKey, adminPassword, and mlServiceUrl
docker compose up --build

That starts the PHP app and the Python worker together, and the admin is at http://localhost:8080/admin.php.

Local without Docker starts the worker yourself, then PHP:

cp config.sample.php config.php
# edit config.php: set apiKey and adminPassword
python3 python/worker.py serve &
php -S localhost:8080 -t public

Step 2: Choose How PHP Reaches Python

One config line, mlDriver, decides how the PHP front talks to the scikit-learn worker. service runs the worker as a tiny HTTP service on localhost, the recommended setup and what Docker wires automatically. cli has PHP run python3 worker.py per request, no daemon to manage, a good fit for light traffic on any box with Python. The main settings:

Step 3: Create a Pipeline and Add Rows

Log into admin.php and create a pipeline. A pipeline holds one or more model steps, and most jobs want exactly one: pick a class, say a KNN classifier for text, and you have a working setup. Choosing a model walks the whole catalog with plain-language advice, and every class shows inline parameter help in the admin.

Then feed it examples. Add rows in the browser, an input and its label, like "refund my order" labeled billing, or push them in bulk through the API. Twenty or thirty rows per label is a fine starting point for a first classifier, and the dataset and training guide covers formats, bulk loading, and dataset habits that pay off.

Step 4: Train and Test

Click Train. The engine rebuilds the model from the complete dataset, vectorizing text and scaling numbers automatically, and reports how many rows it learned from. Then use the test box on the same page: type an input, see the prediction instantly. This tight loop, add rows, train, test, is how models actually get good, and it all happens on one admin page.

Step 5: Call Predict From Your App

Your application talks to one endpoint:

curl -X POST http://localhost:8080/api.php/predict \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_KEY" \
  -d '{"pipelineID": 1, "input": "why was i charged twice"}'

The response is small and direct: {"output": "billing", ...}, plus each step's output when a pipeline has several. Send text for text models or a list of numbers for numeric models, and act on the answer in your code. Every endpoint and field is documented in the API reference.