platform-ai 0.0.1

Prediction API client for Python

prediction-api

• App will receive an HTTP POST call in this format

curl -X POST
    -H 'Authorization: Key API_KEY'
    -H "Content-Type: application/json"
    -d '
    {
      "inputs": [
        {
          "data": {
            "image": {
              "url": "https://lh3.googleusercontent.com/GNE-0Vga-Uffz1i9oxP0ueYYvPW8qG_z7_IlQX3yCqPNILam2Z-rlkYzjoRGkPp74XeEkVb96r1-N1-yRKGhvkFQK5uI=s256"
            }
          }
        }
      ]
    }'
    https://prediction-api-dot-platform-ai-dev.appspot.com/v1/project?id=ahFzfnBsYXRmb3JtLWFpLWRldnJTCxIEVXNlciIcMGFXdExZREhlU1puZ3F1SUxpYXUzQ0NFQ1FFMgwLEgdQcm9qZWN0IiA2MmEwMjYyYjNiYTI0NDA4OGNlYTgyYjg0MTQ1MDg3Mww

• Since this is a multi-tenant application and we cannot redploy it (or use the same key for multi customers), in order to turn on a new tenant, API keys need to be managed in Google Data Store. Project type in Datastore has a reference to valid API keys that can access it for prediction.

• Import model_utilities from platform-ai package (TBD)

  model = get_model(project_id)
model.predict_from_url(URL) returns prediction json

• cache the model in memorystore for future requests, backend will invalidate this cache if/when the model is retrained

Error codes

• Bad URL / not accessible
• project_id invalid
• rate limit exceeded
• no trained model is available
• backend error / something blew up