POT sinkhorn_stabilized and sinkhorn_epsilon_scaling methods for sinkhorn getting stuck #748
Description
Describe the bug
I am running sinkhorn between two datasets of 100 images (each 28x28 -- this is just 2 subsets of fashionmnist), but for some reason, I am finding that the sinkhorn_epsilon_scaling method is getting stuck at a certain value (I have attached the log), no matter if regularization is 0.0001, 0.001, 0.01, 0.1, 1. It is the same for different datasets of the same size -- why is this?
It. |Err
0|8.288824e-25|
10|1.999996e-03|
20|1.999996e-03|
30|1.999996e-03|
40|1.999996e-03|
50|1.999996e-03|
60|1.999996e-03|
70|1.999996e-03|
80|1.999996e-03|
90|1.999996e-03|
It. |Err
100|1.999996e-03|
110|1.999996e-03|
120|1.999996e-03|
130|1.999996e-03|
140|1.999996e-03|
150|1.999996e-03|
160|1.999996e-03|
170|1.999996e-03|
180|1.999996e-03|
190|1.999996e-03|
Environment (please complete the following information):
- OS (e.g. MacOS, Windows, Linux):
- Python version:
- How was POT installed (source,
pip,conda): - Build command you used (if compiling from source):
- Only for GPU related bugs:
- CUDA version:
- GPU models and configuration:
- Any other relevant information:
Output of the following code snippet:
import platform; print(platform.platform())
import sys; print("Python", sys.version)
import numpy; print("NumPy", numpy.__version__)
import scipy; print("SciPy", scipy.__version__)
import ot; print("POT", ot.__version__)