This repository has been archived by the owner on Dec 16, 2022. It is now read-only.
Add a text_to_instance method on DatasetReader
#190
Assignees
Comments
|
I'll do this, should be quick |
|
actually, I take this back, it's not quick, I'll leave it for later |
schmmd
pushed a commit
that referenced
this issue
Feb 26, 2020
* skiff files * move models to json * more stuff * fix cloudbuild.yaml * fix command * change healthcheck * add one model * two services * ingress * only 1replica * change ingress * explicit root path * Set more specific labels on the MC resources. (#189) * parameterize entries * oops * loop over model names * use all the models * add back catchall * try catchall a different way * add more paths * rename ingress * add db info * add cloudsql proxy container * another try * add volumes * fix emptydir * init container * undo init containers * Add an extra bit of information to use as part of the selector. (#190) As it stands the base service includes *all* pods, as they all have the labels it selects against. This adds the `ui_server` label to that pod, so that we won't send traffic to all 13 of them. * match all * Try send_static_file * Moar logs; service labels * Try a catch all route instead. * redirect even when no-models * add back default routes * rename machine-comprehension -> reading-comprehension * allow memory and cpu overrides * handle extra fields correctly * try very few resources * move health check to /health route * better stackdriver logging * maybe this will make logging work correctly * do not propagate * add python-json-logger to Dockerfile * increase time to first healthcheck * address pr comments part 1
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
The
Predictorscurrently have to pull out the tokenizer and the token indexers from theDatasetReaderand recreate what theDatasetReaderdoes internally for every instance. This means that if we want to change (or add parameters to) what theDatasetReaderdoes, we have to change thePredictorsto match. Instead, we should just add atext_to_instancemethod on theDatasetReaderitself, so that thePredictorcan just keep around theDatasetReader, and pass off all processing to it.The text was updated successfully, but these errors were encountered: