Closed ghost closed 1 year ago
Add:
pipeline_components/evaluation/example_count.py
pipeline_components/evaluation/plots_num_feat.html
pipeline_components/evaluation/plots_overall.html
to .gitignore or clean during specific test. These files are created when running make test-all-components
pipeline_components/_tfdv/_tfdv/conftest.py
if empty, is it needed?
Add:
pipeline_components/evaluation/example_count.py pipeline_components/evaluation/plots_num_feat.html pipeline_components/evaluation/plots_overall.html
to .gitignore or clean during specific test. These files are created when running
make test-all-components
Added to .gitignore for this PR. Created #29 to track it and will address properly in a future PR
pipeline_components/_tfdv/_tfdv/conftest.py
if empty, is it needed?
Yes this is required for pytest to pick up the directory so that the contents can be imported during tests
In that case potentially add a comment stating the purpose in the file. Otherwise can image people developing on the template could remove without realising the failing test consequence
In that case potentially add a comment stating the purpose in the file. Otherwise can image people developing on the template could remove without realising the failing test consequence
Added
@Linchin ready for your review :)
Description
The aim of this PR is to separate the Python dependencies for each pipeline component from each other component and from the ML pipelines themselves. There are a lot of files changed (!) but this mostly consists of files moving and some import statements being updated. The key changes are:
pipelines/kfp_components
directory and intopipeline_components
aiplatform
,bigquery
,helpers
etc) has its own Python virtual environment managed withpipenv
and its own tests in its directorypipelines
andtests
directories are moved to a new top-level directory calledpipelines
, which also now contains thePipfile
andPipfile.lock
for the ML pipelines virtual env (previously in the root of the repo)@component
functions directly for use in ML pipelines, the pipeline components are first compiled to YAML (in their respective virtual environment), before being imported from YAML for use in the ML pipelinesHow has this been tested?
Checklist
- [ ] I have added tests that prove my fix is effective or that my feature works