Tfx: Kubeflow TFX on premise examples

[valeriano-manassero]

Hi,
I don't want to spam anything but I just started a repo containing a collection of Kubeflow TFX on premise pipelines: https://github.com/valeriano-manassero/tfx-kubeflow-pipelines

Actually I just have an "Iris" dataset pipeline and a "Cifar-10" one (using Keras model_to_estimator).

I will continue to add and improve pipelines while I'm experimenting them.

There's a chance this repo reference can get included in TFX docs?

Any suggestion appreciated :)

[ruoyu90]

@valeriano-manassero Thanks! Really glad to see those examples. Feel free to send us PRs on those examples and let's discuss which examples to bring in to present new scenarios :)

[valeriano-manassero]

@ruoyu90 since I'm currently working on Kubeflow on premise only, my examples will not cover every use case like taxi pipeline.
I think it'll be better to simply mention my repo in docs, maybe creating a contrib section.
wdyt?

[zhitaoli]

This is great work, Valeriano. Thank you for creating these examples.

On Wed, Sep 4, 2019 at 11:44 PM Valeriano Manassero <
[email protected]> wrote:

@ruoyu90 https://github.com/ruoyu90 since I'm currently working on
Kubeflow on premise only, my examples will not cover every use case like taxi
pipeline.
I think it'll be better to simply mention my repo in docs, maybe creating
a contrib section.
wdyt?

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/tensorflow/tfx/issues/572?email_source=notifications&email_token=AAY6AZSGYRBP5CH6ZHBO5UTQICTFBA5CNFSM4ITFDTL2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD56AX7I#issuecomment-528223229,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAY6AZUSH77MZUVWEXGIOGDQICTFBANCNFSM4ITFDTLQ
.

--
Cheers,

Zhitao Li

[ruoyu90]

@valeriano-manassero I think it's ok to be KFP specific, as long as those examples identify some new patterns (e.g. your cifar-10 example demonstrate a classical image classification problem). We're happy to absorb these examples in :)

[valeriano-manassero]

@ruoyu90 so, I'm going to create a PR asap with probably a MNIST dataset. This is because I can include some data and I will not need to create a custo executor for dowload and this should be better for an easier example.

[valeriano-manassero]

Since PR is there, I think we can close here.