Tensorboard: How to use What-If tool in tensorboard？

[wqw547243068]

I am very excited at the coming tool What-If，but got a little conconfued about how to try it in tensorboard，if anyone knows，please tell me，thanks！

[nfelt]

@jameswex might be able to fill you in. The What-If tool should be available in the next release of TensorBoard. There is a readme available at https://github.com/tensorflow/tensorboard/blob/master/tensorboard/plugins/interactive_inference/README.md

Closing this issue though, since we ask that the issue tracker only be used for reporting bugs and feature requests.

[jameswex]

As of the next release of TensorBoard, you should be able to access the What-If Tool in the dashboards drop-down at the top-right of TensorBoard. Alternatively, you can build from HEAD on master branch, or use a nightly build to see it in TensorBoard. From there, the instructions at https://github.com/tensorflow/tensorboard/blob/master/tensorboard/plugins/interactive_inference/README.md should clarify things. Basically you need a tfrecord file of Example protos and a tensorflow serving (github.com/tensorflow/serving) model server running your model, with your model using either the Classification or Regression API provided by tensorflow serving.

[SameerSharma21]

Hi @jameswex , All the projects which we are working on are DL projects. Also our input datas includes images and time series data. Just wanted to learn from you if what-if tool supports all these ?
Thanks in advance,
Sameer.

[jameswex]

Images as input is supported in the tool and we will put out a colab notebook showing an example soon.

For time series data, you could encode it using tf.SequenceExample protos, which are supported as an input example type in the tool.

[SameerSharma21]

Thanks James. We would be starting to use the tool. Will contact you again if stuck at some place.
Cheers!!