Bug: Getting NotImplemented error for AutoModel fit

[HarshithMjh]

Bug Description

Hi, I am trying to to use AutoModel with custom validation data. The trials are running properly. At the end of all trials, it reaches the code

autokeras/autokeras/engine/tuner.py

Line 243 in db78b44

model.save(self.best_model_path)

Here, I am facing NotImplemented error. I tried using the export_model() and then save with filename parameter. That also gave NotImplemented error.

I saw that the way of getting the model is different for validation_split vs validation_data in the codeblock mentioned above. So, others might not have faced it yet.

Attached reproducible collab link: https://colab.research.google.com/drive/1AQXss_F7L02y3akhJKji2SdmvJxVhnV0?usp=sharing

So, please help with a way where fit completes properly with proper saving. Because, without having the model saved, the whole training will go waste.

Bug Reproduction

Code for reproducing the bug: https://colab.research.google.com/drive/1AQXss_F7L02y3akhJKji2SdmvJxVhnV0?usp=sharing

Data used by the code:

Expected Behavior

AutoModel fir should complete properly even when custom validation data is provided

Setup Details

Include the details about the versions of:

• OS type and version:
• Python:
• autokeras:
• keras-tuner:
• scikit-learn:
• numpy:
• pandas:
• tensorflow:

Additional context

[HarshithMjh]

@haifeng-jin , I understand that you might not be getting spare time to solve it.
But, could you please spend few minutes and help me to get a direction to debug and fix this.
So, I can continue in the direction which you say, and I can only create a PR for the fix