locked python to 3.13 and got rid of dumb build support

Author: pavelkomarov

.github/workflows/test.yml

@@ -13,13 +13,9 @@ jobs:
           fetch-depth: 0 # important for Coveralls to read commit history
       - uses: actions/setup-python@v5
         with:
-          python-version: '3.x'
+          python-version: '3.13' # because cvxopt binaries for 3.14 don't yet exist, and building is terrible
       - name: tests and coverage
-        run: | # lines before pip install are for building cvxopt from source, which is recently necessary for Python 3.14, too new for prebuilt wheel
-          sudo apt-get install -y libblas-dev liblapack-dev libsuitesparse-dev libglpk-dev
-          export CVXOPT_GLPK=True
-          export CVXOPT_GLPK_INC=/usr/include
-          export CVXOPT_GLPK_LIB=/usr/lib/x86_64-linux-gnu
+        run: |
           pip install -e .[advanced,dev] coveralls
           coverage run --source=pynumdiff --omit='pynumdiff/_version.py,pynumdiff/tests/*,pynumdiff/utils/old_pi_cruise_control.py' -m pytest -s
           coverage xml

pynumdiff/optimize/_optimize.py

@@ -90,11 +90,12 @@
     robustdiff: ({'order': {1, 2, 3}, # warning: order 1 hacks the loss function when tvgamma is used, tends to win but is usually suboptimal choice in terms of true RMSE
                       'q': [1e-1, 1e1, 1e4, 1e8, 1e12],
                       'r': [1e-1, 1e1, 1e4, 1e8, 1e12],
-            'proc_huberM': {0, 1, 1.345, 2, 6}, # 0 is l1 norm, 1.345 is Huber 95% "efficiency", 2 assumes about 5% outliers,
-            'meas_huberM': {0, 1, 1.345, 2, 6}}, # and 6 assumes basically no outliers -> l2 norm. Try (1 - norm.cdf(M))*2 to see outlier portion
+            'proc_huberM': [0., 2, 6], # 0 is l1 norm, 1.345 is Huber 95% "efficiency", 2 assumes about 5% outliers,
+            'meas_huberM': [0., 2, 6]}, # and 6 assumes basically no outliers -> l2 norm. Try (1 - norm.cdf(M))*2 to see outlier portion
                      {'q': (1e-1, 1e18),
                       'r': (1e-5, 1e18),
-                 'huberM': (0, 5)}), # really only want to use l2 norm when nearby
+            'proc_huberM': (0, 6),
+            'meas_huberM': (0, 6)}),
     lineardiff: ({'kernel': 'gaussian',
                    'order': 3,
                    'gamma': [1e-1, 1, 10, 100],