ValueError: Digitalized labels does not have elements close enough to bin index 1. The bin index should be in the range of the labels values.

[FrancescaGuo]

I have been struggling with this value error when I use the Consistency demo codes on my own mice behavioral data. Here are a few screenshots for my codes and a full description of the error. I have trained the Cebra_Behavior model with mice neural data and behavioral data. I successfully get the embeddings and visualize them but get stuck when I try to calculate the consistency of the embeddings between mice.

'Mouselabel' is a 2D list which has the same shape as my embeddings. 'Names' is a list of integers with the mouse ID (1 to 14).


I would really appreciate it if someone has any clue of what I did wrong.

[stes]

Hi @FrancescaGuo , apologies for the late response here. Please also see my related reply here: #125 (comment)

The code uses the default number of bins (100) and partitions the behavioral data into this number of bins. In your example, it would be a reasonable choice to set the number of bins to the number of discrete labels (14).

[stes]

I created an issue to improve the code here: #128

In the meantime, please consider setting n_bins manually to the number of discrete labels. Let me know if this works!

[stes]

Hello @FrancescaGuo , just checking, were you able to fix the issue by adapting the bin size?

[ELGarulli]

Hi @stes I was getting the same error as @FrancescaGuo and setting the num_discretization_bins to the unique values in my labels solved it.
I just want to mention that at first I was confused when you suggested "number of discrete labels (14)", as I understand OP has 14 single session models to compare, but the number of values of labels could be any other number, is this correct?
I am adding this note as I would like to check if I misunderstood something. I am trying to train a multi-session model. Each session is a separate run (s=55), my labels are a list of arrays each as long as the corresponding run, each label array can contain max 3 values (0, 1, 2).
So for me setting the num_discretization_bins to 3 is what solves the "Digitalized labels does not have elements close enough to bin" error.