Google Cloud Platform (GCP) AutoML Products Overview
Google Cloud Platform (GCP)
Codeless model building, built with Neural Architecture Search (NAS) & using NASNets
Products
Cloud AutoML Natural Language
Cloud AutoML Translation
Cloud AutoML Vision
Features
TRAINING: Uses a Prepared Dataset to train a Custom Model. 1 Prepared Dataset = 1 Custome Model
Make a small Prepared Dataset for experimentation in the UI
Assemble data in a CSV file, UTF-8 encoded; Stored in same Cloud Storage Bucket with the source files; no blank or duplicate lines; no Unicode characters
CSV Column #1 is optional - for TRAIN, VALIDATION, TEST. If blank, rows will be assigned 80% TRAIN, 10% VALIDATION, 10% TEST
CSV Column #2 - identifies source files in Cloud Storage; can be ZIP
CSV Column #3+ - labels; alphanumeric with underscores, no special characters
Can create in code: Python, Java, Node.js
Do NOT expire
Custom Model
Temporary - eventually deleted after a period - so you must periodically train a new model
Cannot be exported or saved externally
NodeLabel
DEPLOYMENT
https://automl.googleapis.com/
Cold starts if model hasn't been used for a while - possibly minutes delay
SERVING
Request contains the model-name and payload
Response contains JSON with multiple label fields named 'displayName', the keyword 'classification' and 'score' confidence level 0 to 1.0
Quotas apply to model creation and service requests