Runtime error when using LLM_abba: AttributeError / NoneType object issues during fit and transform stages

[Fly-Cat-0]

Hi, thanks for the great work on this repository!

I'm currently running the demo-ts-classification.py example script for time series classification, and I encountered an error during the inference phase. Below are the details:
Traceback (most recent call last):
File "/home/zengjie/llm-abba-0.0.4/examples/demo-ts-classification.py", line 70, in
out_text = LLMABBA_classification.inference(
File "/home/zengjie/llm-abba-0.0.4/llmabba/llmabba.py", line 560, in inference
symbols_test_data, params_test_data = self.xabba.transform(data_scaled)
File "/home/zengjie/llm-abba-0.0.4/llmabba/xabba.py", line 695, in transform
series = series.reshape(-1, int(np.prod(self.recap_shape[1:])))
TypeError: 'NoneType' object is not subscriptable
Possible Cause:
From tracing the code, it seems that self.recap_shape is still None during transform(). This likely means that fit() was never called on xabba before transform(). But in the inference() method of LLMABBA, it directly calls self.xabba.transform(...) without any prior fitting.
My Questions:
Should self.xabba.fit(data_scaled) be explicitly called inside inference() before calling .transform()?
Or is it expected that users manually call fit() on the model before running inference?
If so, can we update the example script to reflect this?
Looking forward to your advice or a possible fix. Thanks again!
Best,
Fly-Cat-0

[chenxinye]

Hi @Fly-Cat-0
Thanks for the issue, I will get back to you as soon as possible!

Best,
Xinye

[Fly-Cat-0]

Hi @xinye-chen,

Thanks for your prompt response!

I looked into the issue and found that the problem lies in the XABBA class within xabba.py. Specifically, the self.recap_shape variable is initially set to None, but in cases where the input data has less than 3 dimensions, it is never reassigned.

After explicitly assigning self.recap_shape = input_data.shape in that branch, the code runs smoothly without errors.

Just wanted to share this in case it helps with debugging or future improvements.

Best regards,
Fly-Cat-0

[chenxinye]

@Fly-Cat-0 Thanks for your informative feedback. I think you didn't fit first before you run the inference, right? can you show me your original run example, do you follows the README.md guidance?

[Fly-Cat-0]

@chenxinye I ran the classification script provided in the code/examples directory directly, and I also tried the script from the README.md.
However, I didn’t see any instructions in the README.md mentioning the need to call fit() before running inference.

[chenxinye]

@Fly-Cat-0 Thanks for your feedback, I mean did you call the train method first?

[chenxinye]

It requires to fit the data first, I believe @ChengKang520 can better answer your question.

[ChengKang520]

@chenxinye I ran the classification script provided in the code/examples directory directly, and I also tried the script from the README.md. However, I didn’t see any instructions in the README.md mentioning the need to call fit() before running inference.

Hi, thanks. Sorry, I did not realise that you were running the example. When you are doing the test, you need to fit the input data through the saved ABBA.pkl file. That means you need to transform the input time series into a symbolic series under the same Scaler. Please see line 541 of llmabba/llmabba.py

Best,

Cheng

[ChengKang520]

@chenxinye I ran the classification script provided in the code/examples directory directly, and I also tried the script from the README.md. However, I didn’t see any instructions in the README.md mentioning the need to call fit() before running inference.

There are some tips that can help your project.

For example, if you are doing the PTBDB classification task, design the instruction as below:

###Instruction: "You will be given an instruction, domain, task, inputs and outputs. Your objective is to evaluate the input and complete the task by the information presented in the evidence.\n"
###Domain:"ECG abnormality"
###Task: "This is a classification task. Identify the 'ECG Abnormality' according to the given 'Symbolic Series'."
###Evidence: An ECG abnormality, also known as an abnormal electrocardiogram, has a chain of logic that can be represented by a symbolic series.
###Input: {*symbolic series}
###Output: {*lables}

Please carefully set abba_tol and abba_alpha to avoid too long input.