Add cython-lint pre-commit hook (#623)

.pre-commit-config.yaml

@@ -41,3 +41,9 @@ repos:
    hooks:
     - id: pyupgrade-conda
       exclude: ^src/glum_benchmarks/orig_sklearn_fork/
+ - repo: https://github.com/Quantco/pre-commit-mirrors-cython-lint
+   rev: 0.14.0
+   hooks:
+    - id: cython-lint-conda
+      args: [--no-pycodestyle]
+    - id: double-quote-cython-strings-conda

src/glum/_cd_fast.pyx

@@ -13,7 +13,6 @@ cimport numpy as np
 import numpy as np
 from numpy.math cimport INFINITY
 
-cimport cython
 from cython cimport floating
 from cython.parallel import prange
 import warnings
@@ -55,6 +54,7 @@ cdef inline floating fsign(floating f) nogil:
     else:
         return -1.0
 
+
 def identify_active_rows(
     floating[::1] gradient_rows,
     floating[::1] hessian_rows,
@@ -89,6 +89,7 @@ def identify_active_rows(
 
     return hessian_rows_diff_arr, active_rows_arr
 
+
 def enet_coordinate_descent_gram(int[::1] active_set,
                                  floating[::1] w,
                                  floating[::1] P1,
@@ -111,7 +112,6 @@ def enet_coordinate_descent_gram(int[::1] active_set,
 
     # get the data information into easy vars
     cdef unsigned int n_active_features = active_set.shape[0]
-    cdef unsigned int n_features = Q.shape[0]
 
     cdef floating w_ii
     cdef floating P1_ii
@@ -197,6 +197,7 @@ def enet_coordinate_descent_gram(int[::1] active_set,
 
     return np.asarray(w), norm_min_subgrad, max_min_subgrad, tol, n_iter + 1
 
+
 cdef void cython_norm_min_subgrad(
     int[::1] active_set,
     floating[::1] coef,
@@ -268,6 +269,7 @@ cdef void cython_norm_min_subgrad(
         if absterm > max_out[0]:
             max_out[0] = absterm
 
+
 def _norm_min_subgrad(
     int[::1] active_set,
     floating[::1] coef,

src/glum/_functions.pyx

@@ -20,6 +20,7 @@ cdef fused const_floating1d:
     const float[:]
     const double[:]
 
+
 # NOTE: Here and below, the factor argument is left for last. That's because it
 # will be a single floating point value being passed from Python. In Python,
 # all floating point values are 64-bit. As a result, if it's the first
@@ -38,7 +39,6 @@ def normal_identity_eta_mu_deviance(
 ):
     cdef int n = cur_eta.shape[0]
     cdef int i
-    cdef floating unit_deviance
     cdef floating deviance = 0.0
     for i in prange(n, nogil=True):
         eta_out[i] = cur_eta[i] + factor * X_dot_d[i]
@@ -49,6 +49,7 @@ def normal_identity_eta_mu_deviance(
         deviance += weights[i] * (y[i] - mu_out[i]) ** 2
     return deviance
 
+
 def normal_identity_rowwise_gradient_hessian(
     const_floating1d y,
     const_floating1d weights,
@@ -65,6 +66,7 @@ def normal_identity_rowwise_gradient_hessian(
         # the default calculation in _distribution.py
         hessian_rows_out[i] = weights[i]
 
+
 def normal_log_likelihood(
     const_floating1d y,
     const_floating1d weights,
@@ -110,7 +112,6 @@ def poisson_log_eta_mu_deviance(
 ):
     cdef int n = cur_eta.shape[0]
     cdef int i
-    cdef floating unit_deviance
     cdef floating deviance = 0.0
     for i in prange(n, nogil=True):
         eta_out[i] = cur_eta[i] + factor * X_dot_d[i]
@@ -180,7 +181,6 @@ def gamma_log_eta_mu_deviance(
 ):
     cdef int n = cur_eta.shape[0]
     cdef int i
-    cdef floating unit_deviance
     cdef floating deviance = 0.0
     for i in prange(n, nogil=True):
         eta_out[i] = cur_eta[i] + factor * X_dot_d[i]
@@ -252,7 +252,6 @@ def tweedie_log_eta_mu_deviance(
 ):
     cdef int n = cur_eta.shape[0]
     cdef int i
-    cdef floating unit_deviance
     cdef floating deviance = 0.0
     cdef floating mu1mp
     for i in prange(n, nogil=True):