[Question] Project Algorithms Milestones and its relations to Kerastuner

[Uiuran]

Hi,

You seem to be using RandomSearch and params band search (HyperBand) from kerastuner as algorithms. You are planning to deploy other popular NAS/AutoML as Reinforcement Learning and Meta-Heuristic Search (GA e.g.) ? Does keras-team plans to do all the algorithms in kerastuner or in this module ? Do you have a road to the milestones of it until the beta ?

NAS is a very new research topic and seems to be very exciting to build a module to easy the flow through this research ...

To note: master branch is not specifying kerastuner as dependency as-far-i-have-seem.

thank you

[tuokri]

Commenting here as this is related to Keras Tuner not being specified as a dependency as mentioned in the OP.

Installing autokeras from source (master branch) results in the following error upon import:

>>> import autokeras
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/tuokri/autokeras/autokeras/__init__.py", line 1, in <module>
    from autokeras.auto_model import AutoModel
  File "/home/tuokri/autokeras/autokeras/auto_model.py", line 1, in <module>
    import kerastuner
ModuleNotFoundError: No module named 'kerastuner'

This is fixed by manually installing Keras Tuner: pip install keras-tuner

[bhack]

I ask also for autoaugmentation plans. See https://github.com/keras-team/keras-tuner/issues/153

[bhack]

Something it Is landing in TPU https://arxiv.org/abs/1909.13719 https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/autoaugment.py

[haifeng-jin]

The current dependency of kerastuner is based on rc0. You can install kerastuner with the following command. pip install git+https://github.com/keras-team/keras-tuner.git@1.0rc0. You can always check the kerastuner version we are depending on in .travis.yml.

[Uiuran]

Adding as idea of project milestone, or something i didnt see anywhere else but seems significant for research in AutoML/NAS, and maybe is good to think in an implementation, are the Monte Carlo Tree Search models to architecture search.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.