Speedup import time with lazy import of scipy.stats

[ricardoV94]

This reduces pytensor import time from average 0.9 to 0.45s on my machine

Before:

python -m benchmark_imports pytensor 0.8324 root pytensor 0.3692 project pytensor.tensor.random 0.3689 project pytensor.tensor.random.rewriting 0.3649 project pytensor.tensor.random.rewriting.jax 0.3640 project pytensor.tensor.random.basic 0.3616 dependency scipy.stats 0.2857 transitive scipy.stats._stats_py from scipy.stats 0.2108 project pytensor.scalar 0.1785 project pytensor.scalar.math 0.1615 dependency scipy.special 0.1484 transitive scipy.stats.distributions from scipy.stats._stats_py 0.1056 project pytensor.configdefaults 0.1020 transitive scipy.stats._continuous_distns from scipy.stats.distributions 0.0888 dependency numpy 0.0794 transitive scipy.special._support_alternative_backends from scipy.special 0.0791 transitive scipy._lib._array_api from scipy.special._support_alternative_backends 0.0767 transitive scipy._lib.array_api_compat.numpy from scipy._lib._array_api 0.0760 project pytensor.tensor 0.0711 transitive scipy.optimize from scipy.stats._stats_py 0.0656 transitive scipy.special._orthogonal from scipy.special 0.0649 transitive scipy.linalg from scipy.special._orthogonal 0.0645 project pytensor.tensor.rewriting 0.0477 transitive scipy.linalg._sketches from scipy.linalg 0.0474 transitive numpy.f2py from scipy._lib.array_api_compat.numpy 0.0470 transitive numpy.f2py.f2py2e from numpy.f2py 

After:

python -m benchmark_imports pytensor 0.4396 root pytensor 0.1776 project pytensor.scalar 0.1490 project pytensor.scalar.math 0.1348 dependency scipy.special 0.0836 project pytensor.configdefaults 0.0766 project pytensor.tensor 0.0733 transitive scipy.special._support_alternative_backends from scipy.special 0.0730 transitive scipy._lib._array_api from scipy.special._support_alternative_backends 0.0708 transitive scipy._lib.array_api_compat.numpy from scipy._lib._array_api 0.0670 project pytensor.tensor.rewriting 0.0647 dependency numpy 0.0466 transitive scipy.special._orthogonal from scipy.special 0.0460 transitive scipy.linalg from scipy.special._orthogonal 0.0417 transitive numpy.f2py from scipy._lib.array_api_compat.numpy 0.0413 transitive numpy.f2py.f2py2e from numpy.f2py 0.0312 transitive scipy.linalg._sketches from scipy.linalg 0.0305 transitive scipy.sparse from scipy.linalg._sketches 0.0302 transitive numpy.__config__ from numpy 0.0297 transitive numpy._core from numpy.__config__ 0.0290 transitive numpy.f2py.crackfortran from numpy.f2py.f2py2e 0.0282 project pytensor.scalar.basic 0.0262 project pytensor.tensor.rewriting.basic 0.0244 project pytensor.tensor.random 0.0242 transitive numpy.lib from numpy 0.0241 project pytensor.tensor.random.rewriting 

---

📚 Documentation preview 📚: https://pytensor--1268.org.readthedocs.build/en/1268/

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 81.99%. Comparing base (bf628c9) to head (5df5bdc).
Report is 145 commits behind head on main.

Additional details and impacted files

@@ Coverage Diff @@ ## main #1268 +/- ## ======================================= Coverage 81.99% 81.99% ======================================= Files 188 188 Lines 48608 48610 +2 Branches 8688 8689 +1 ======================================= + Hits 39857 39859 +2  Misses 6586 6586 Partials 2165 2165 

Files with missing lines	Coverage Δ
pytensor/tensor/random/basic.py	98.86% <100.00%> (+<0.01%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.