Draw pr curve with provided raw data.

[foreverYoungGitHub]

Hi,

Thanks so much for you to provide so convenient tools to visualize the process for pytorch.

Recently, I want draw the pr curve and other curves with tensorboard by directly provided the precision and recall values. However, I found that the inputs of add_pr_curve is the predict value and ground truth. It calculated the pr curve by using compute_curve().

Is that possible that add a new interface like add_pr_curve_raw to add pr curve officially by provided the raw data. Just like pr_curve_raw_data_pb in tensorboard.

Thanks so much!

Best wishes,

[lanpa]

No problem.

[foreverYoungGitHub]

Thanks so much!

[lanpa]

@foreverYoungGitHub Have a try~

[foreverYoungGitHub]

Thank you for your help!

I draw some pr_curves and other curves like this:

Thanks for your help!

Besides, do you have any plan to add something like custom_scalar?

Thanks again for your work!

Best wishes,

[lanpa]

So does the result make sense? custom_scalar is cool, I will bring it to tensorboardX.

[foreverYoungGitHub]

It is great to hear that!

The current result is great! Just some curve does not belong to PR curve and it has to normalize into 0 to 1. But it's some minor problem.

[MaximArtemev]

Hi!
Any success with bringing custom_scalar into tensorboardX?

[lanpa]

@MaximArtemev Haven't start yet. 😂

[zhanghanbin3159]

@foreverYoungGitHub how do you draw pr-curve with the precision and recall values? the add_pr_curve_raw function ?

[foreverYoungGitHub]

@zhanghanbin3159 please check this code