Tutorial 00, objective error_metric vs error_squared_norm?

[DanielTakeshi]

Has there been a recent change that deprecated error_metric() in favor of error_squared_norm()? I'm running tutorial 00 for Theseus, from today (as of main branch) and am running into this issue when I try and run the "Objective" category.

For ease of copy and paste, the error text is below:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[6], line 35
     33 print(f"Error term 2: unweighted: {cf2.error()} weighted: {cf2.weighted_error()}")
     34 # Objective value should be: (error1)^2 + (error2)^2 
---> 35 print(f"Objective value: {objective.error_metric()}")

AttributeError: 'Objective' object has no attribute 'error_metric'

However, switching it to use error_squared_norm() seems to be fine:

This value of 4.250 also matches the one that I now see on the public GitHub:

https://github.com/facebookresearch/theseus/blob/main/tutorials/00_introduction.ipynb

The relevant part is located here and uses error_metric():

Has there been a recent change which deprecated some of the APIs? If so, might be worth updating the notebook examples.

[luisenp]

Ah, the error_metric method was introduced recently, after version 0.1.4. The tutorials are consistent with the latest source code, but I missed that updating tutorials with this new change would break them for older versions of theseus. Sorry about this.

Switching to error_squared_norm() is the correct thing to do if you are on an older version.

[DanielTakeshi]

Got it, thanks! I think PyPI still uses 0.1.4 https://pypi.org/project/theseus-ai/ but presumably that will get updated sometime in the future.

[luisenp]

Yes. Should be updated to 0.2.0 by the end of this month.