Analyzer caching is disabled at the moment since no artifact exists in the component spec to give a working directory to write cache information. After changes, a new TransformCache artifact type is created, and the Transform component optionally takes in a input channel of type TransformCache (if cache exists) and, in addition to current outputs, also outputs a TransformCache output artifact, where caching information is stored.
Since optional output artifacts are not yet possible and since caching information is calculated regardless of whether a cache output artifact to write to is provided, current behavior is to use a input cache artifact if provided, and to always provide a cache output artifact.
该提问来源于开源项目：tensorflow/tfx
