Artificial Intelligence Recommendation:Set up a PAI-Rec project

Prerequisites

Before you begin, make sure you have:

• Go installed and GOPATH configured

• Docker installed (required to build and push container images)

• An Alibaba Cloud Container Registry instance (required for the image packaging step)

Download pairecmd

Download the binary for your platform:

On Linux or macOS, grant execute permission after downloading. Replace <binary> with the downloaded filename:

chmod +x <binary>

Version 2 of pairecmd supports the PAI-Rec console. For production projects, use the console to manage your project configuration.

Create a project

Run the following command to create a project named pairec-demo. The command generates a pairec-demo folder containing the project scaffold in the current directory.

./pairecmdmac project --name pairec-demo

The generated folder has the following structure:

pairec-demo
├── Makefile
├── conf
│   └── config.json.production
├── docker
│   └── Dockerfile
├── go.mod
└── src
    ├── controller
    │   └── feed.go
    └── main.go

Navigate into the project folder and install dependencies:

cd pairec-demo
go mod tidy

Compile and package the project

The project includes a Makefile with targets for compiling and packaging.

Compile the project:

make && make build

If the compilation fails, run go mod tidy first to resolve any missing dependencies.

Package the project into a container image:

make release

Before running make release, activate a personal image in Alibaba Cloud Container Registry and create the image repository referenced by BIN_NAME in the Makefile.

Run the project

Navigate to the pairec-demo folder and start the service. Use the config parameter to specify the configuration file path. Logs are printed to the terminal.

go run src/main.go --config=conf/config.json.production --alsologtostderr

To verify the service is running, open another terminal and send a test request:

curl -v http://localhost:8000/api/rec/feed -d '{"uid":"76295990", "size":10, "scene_id":"home_feed"}'

A successful response returns HTTP 200 with a JSON body similar to the example in the Request and response example section below.

For details on the config parameter options, see Engine configuration.

API reference

The generated project includes a built-in recommendation API. The implementation is in controller/feed.go.

POST /api/rec/feed

curl -v http://host/api/rec/feed -d '{"uid":"76295990", "size":10, "scene_id":"home_feed"}'

{
    "code": 200,
    "msg": "success",
    "request_id": "e332fe9c-7d99-45a8-a047-bc7ec33d07f6",
    "size": 10,
    "experiment_id": "",
    "items": [
        {
            "item_id": "248791390",
            "score": 0.9991594902203332,
            "retrieve_id": "mock_recall"
        }
    ]
}

"SceneConfs": {
  "${scene_name}": {
    "default": {
      "RecallNames": [
        "collaborative_filter"
      ],
      "OutputFields": [
        "item:type",
        "item:age",
        "score:*",
        "score:model_v1_ctr"
      ]
    }
  }
}

FAQ

How do I pass an item list through the API for reranking?

Pass the list using the item_list parameter. In the engine configuration, set up a Context Item Recall (ContextItemRecall) so the PAI-Rec engine reads items directly from item_list.

How do I instrument the client to use experiment reports?

The API response includes request_id and experiment_id. Record both fields in your impression and click behavior logs at a minimum. Including them in additional logs — such as playback duration, add-to-cart events, and purchases — gives more complete performance data.

Once your behavior logs are instrumented, generate the offline Experiment Report Source Table and register it with the AB Test platform. Configure custom metrics as needed, then view results in the Experiment Metrics Report. For the SQL pattern used to generate the source table, see Data Registration and Field Configuration.

What's next

• Engine configuration — configure recall, ranking, and filtering pipelines

• Data Registration and Field Configuration — register user and item data

• Experiment Metrics Report — analyze A/B testing results