Fix pytorch versions ci (#3289)

* Fixed torch.pi usage unavailable in pytorch 1.5.1, 1.8.1

* Fix pytorch versions CI failures

Author: vfdev-5

Diff for: .github/workflows/pytorch-version-tests.yml

@@ -19,13 +19,8 @@ jobs:
         # will drop python version and related pytorch versions
         python-version: [3.8, 3.9, "3.10"]
         pytorch-version:
-          [2.3.1, 2.2.2, 2.1.2, 2.0.1, 1.13.1, 1.12.1, 1.10.0, 1.8.1, 1.5.1]
+          [2.3.1, 2.2.2, 2.1.2, 2.0.1, 1.13.1, 1.12.1, 1.10.0, 1.8.1]
         exclude:
-          - pytorch-version: 1.5.1
-            python-version: 3.9
-          - pytorch-version: 1.5.1
-            python-version: "3.10"
-
           # disabling python 3.9 support with PyTorch 1.7.1 and 1.8.1, to stop repeated pytorch-version test fail.
           # https://github.com/pytorch/ignite/issues/2383
           - pytorch-version: 1.8.1
@@ -74,6 +69,13 @@ jobs:
         shell: bash -l {0}
         run: |
           conda install pytorch=${{ matrix.pytorch-version }} torchvision cpuonly python=${{ matrix.python-version }} -c pytorch
+          
+          # We should install numpy<2.0 for pytorch<2.3
+          numpy_one_pth_version=$(python -c "import torch; print(float('.'.join(torch.__version__.split('.')[:2])) < 2.3)")
+          if [ "${numpy_one_pth_version}" == "True" ]; then
+            pip install -U "numpy<2.0"
+          fi
+
           pip install -r requirements-dev.txt
           python setup.py install
 

Diff for: ignite/metrics/hsic.py

@@ -134,6 +134,7 @@ def update(self, output: Sequence[Tensor]) -> None:
 
         vx: Union[Tensor, float]
         if self.sigma_x < 0:
+            # vx = torch.quantile(dxx, 0.5)
             vx = torch.quantile(dxx, 0.5)
         else:
             vx = self.sigma_x**2

Diff for: tests/ignite/metrics/test_hsic.py

@@ -1,3 +1,4 @@
+import math
 from typing import Tuple
 
 import numpy as np
@@ -74,7 +75,7 @@ def test_case(request) -> Tuple[Tensor, Tensor, int]:
         N = 200
         b = 20
         x = torch.randn(N, 5)
-        y = x @ torch.normal(0.0, torch.pi, size=(5, 3))
+        y = x @ torch.normal(0.0, math.pi, size=(5, 3))
         y = (
             torch.stack([torch.sin(y[:, 0]), torch.cos(y[:, 1]), torch.exp(y[:, 2])], dim=1)
             + torch.randn_like(y) * 1e-4