removed scipy from build-only dependencies (#3538)

* removed scipy from build-only dependencies; updated README accordingly

* fixed conditions regarding oldest numpy installation on arm64 Windows machines

* README rephrased regarding NumPy installation

* Revert "fixed conditions regarding oldest numpy installation on arm64 Windows machines"

This reverts commit 68197fb.

* pin scipy, need old sparsetools for wheel testing

* make scipy pin more strict

* use macos-latest instead of deprecated macos-11

* pin scipy in setup.py

* remove redundant scipy pin from build-wheels.yml

* get rid of remaining macos-11 traces in CI workflow

* use macos-12 instead of macos-latest

---------

Co-authored-by: Michael Penkov <m@penkov.dev>

.github/workflows/build-wheels.yml

@@ -23,7 +23,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-20.04, windows-2019, macos-11]
+        os: [ubuntu-20.04, windows-2019, macos-12]
     steps:
 
       - name: Checkout
@@ -61,11 +61,11 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - {python: '3.8', os: macos-11}
-          - {python: '3.9', os: macos-11}
-          - {python: '3.10', os: macos-11}
-          - {python: '3.11', os: macos-11}
-          - {python: '3.12', os: macos-11}
+          - {python: '3.8', os: macos-12}
+          - {python: '3.9', os: macos-12}
+          - {python: '3.10', os: macos-12}
+          - {python: '3.11', os: macos-12}
+          - {python: '3.12', os: macos-12}
 
           - {python: '3.8', os: ubuntu-20.04}
           - {python: '3.9', os: ubuntu-20.04}

README.md

@@ -49,13 +49,12 @@ on Wikipedia.
 Installation
 ------------
 
-This software depends on [NumPy and Scipy], two Python packages for
-scientific computing. You must have them installed prior to installing
-gensim.
-
-It is also recommended you install a fast BLAS library before installing
-NumPy. This is optional, but using an optimized BLAS such as MKL, [ATLAS] or
-[OpenBLAS] is known to improve performance by as much as an order of
+This software depends on [NumPy], a Python package for
+scientific computing. Please bear in mind that building NumPy from source
+(e.g. by installing gensim on a platform which lacks NumPy .whl distribution)
+is a non-trivial task involving [linking NumPy to a BLAS library].  
+It is recommended to provide a fast one (such as MKL, [ATLAS] or
+[OpenBLAS]) which can improve performance by as much as an order of
 magnitude. On OSX, NumPy picks up its vecLib BLAS automatically,
 so you don’t need to do anything special.
 
@@ -69,7 +68,9 @@ Or, if you have instead downloaded and unzipped the [source tar.gz]
 package:
 
 ```bash
-    python setup.py install
+    tar -xvzf gensim-X.X.X.tar.gz
+    cd gensim-X.X.X/
+    pip install .
 ```
 
 For alternative modes of installation, see the [documentation].
@@ -172,7 +173,8 @@ BibTeX entry:
   [documentation and Jupyter Notebook tutorials]: https://github.com/RaRe-Technologies/gensim/#documentation
   [Vector Space Model]: https://en.wikipedia.org/wiki/Vector_space_model
   [unsupervised document analysis]: https://en.wikipedia.org/wiki/Latent_semantic_indexing
-  [NumPy and Scipy]: https://scipy.org/install/
+  [NumPy]: https://numpy.org/install/
+  [linking NumPy to a BLAS library]: https://numpy.org/devdocs/building/blas_lapack.html
   [ATLAS]: https://math-atlas.sourceforge.net/
   [OpenBLAS]: https://xianyi.github.io/OpenBLAS/
   [source tar.gz]: https://pypi.org/project/gensim/

pyproject.toml

@@ -9,10 +9,6 @@ requires = [
     # is 1.18.5, remove the line when they increase oldest supported Numpy for this platform
     "numpy==1.18.5; python_version=='3.8' and platform_machine not in 'arm64|aarch64'",
     "oldest-supported-numpy; python_version>'3.8' or platform_machine in 'arm64|aarch64'",
-    #
-    # SciPy 1.14.0 onwards removed the deprecated sparsetools submodule needed by gensim.models.lsimodel
-    #
-    "scipy<=1.13.1",
     "setuptools",
     "wheel",
 ]

setup.py

@@ -331,7 +331,10 @@ def run(self):
 
 install_requires = [
     NUMPY_STR,
-    'scipy >= 1.7.0',
+    #
+    # scipy 1.14.0 and onwards removes deprecated sparsetools submodule
+    #
+    'scipy >= 1.7.0, <1.14.0',
     'smart_open >= 1.8.1',
 ]