Merge pull request #30 from Transurgeon/main

scale benchmarks so they are a bit more stable

Author: PTNobel

.github/workflows/build.yml

@@ -22,10 +22,10 @@ jobs:
   tests:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
-      - uses: actions/setup-python@v2
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v4
         with:
-          python-version: 3.8
+          python-version: "3.12"
       - name: install and test
         run: |
           pip install .

backends/dpp_canonicalization.py

@@ -42,9 +42,8 @@
 from dataclasses import dataclass, field
 from typing import Callable
 
-import numpy as np
-
 import cvxpy as cp
+import numpy as np
 
 # Optional imports
 try:
@@ -1184,8 +1183,12 @@ def main():
         profile_problems = {
             "dpp_lasso_small": ("DPP LASSO (50x100)", make_dpp_lasso(50, 100)),
             "dpp_lasso_medium": ("DPP LASSO (200x500)", make_dpp_lasso(200, 500)),
-            "dpp_constraint_medium": ("DPP Constraint (500x200)", make_dpp_constraint_matrix(200, 500)),
-            "dpp_constraint_large": ("DPP Constraint (1000x500)", make_dpp_constraint_matrix(500, 1000)),
+            "dpp_constraint_medium": (
+                "DPP Constraint (500x200)", make_dpp_constraint_matrix(200, 500)
+            ),
+            "dpp_constraint_large": (
+                "DPP Constraint (1000x500)", make_dpp_constraint_matrix(500, 1000)
+            ),
         }
         if args.profile in profile_problems:
             name, factory = profile_problems[args.profile]

benchmark/finance.py

@@ -84,7 +84,7 @@ def time_compile_problem(self):
 
 class FactorCovarianceModel:
     def setup(self):
-        n = 14500
+        n = 25000
         m = 250
         np.random.seed(1)
         mu = np.abs(np.random.randn(n, 1))

benchmark/gini_portfolio.py

@@ -20,7 +20,7 @@ class Yitzhaki:
     def setup(self):
         rs = np.random.RandomState(123)
         N = 50
-        T = 150
+        T = 300
         cov = rs.rand(N, N) * 1.5 - 0.5
         cov = cov @ cov.T/1000 + np.diag(rs.rand(N) * 0.7 + 0.3)/1000
         mean = np.zeros(N) + 1/1000
@@ -51,7 +51,7 @@ class Murray:
     def setup(self):
         rs = np.random.RandomState(123)
         N = 50
-        T = 350
+        T = 700
         cov = rs.rand(N, N) * 1.5 - 0.5
         cov = cov @ cov.T/1000 + np.diag(rs.rand(N) * 0.7 + 0.3)/1000
         mean = np.zeros(N) + 1/1000
@@ -95,7 +95,7 @@ class Cajas:
     def setup(self):
         rs = np.random.RandomState(123)
         N = 50
-        T = 350
+        T = 800
         cov = rs.rand(N,N) * 1.5 - 0.5
         cov = cov @ cov.T/1000 + np.diag(rs.rand(N) * 0.7 + 0.3)/1000
         mean = np.zeros(N) + 1/1000

benchmark/high_dim_convex_plasticity.py

@@ -62,7 +62,7 @@ def criterion(sig: cp.Variable, p: cp.Variable):
         )
         deps_dummy = np.full((4000, 4), particular_deps)
 
-        N = 500
+        N = 3000
         deps_local_values = deps_dummy[:N]
 
         deps = cp.Parameter((4, N), name="deps")

benchmark/matrix_stuffing.py

@@ -19,8 +19,8 @@
 class ConeMatrixStuffingBench:
 
     def setup(self):
-        m = 2000
-        n = 2000
+        m = 5000
+        n = 5000
         A = np.random.randn(m, n)
         C = np.random.rand(m // 2)
         b = np.random.randn(m)
@@ -68,8 +68,8 @@ def time_compile_problem(self):
 class SmallMatrixStuffing:
 
     def setup(self):
-        m = 200
-        n = 200
+        m = 4000
+        n = 4000
         A = np.random.randn(m, n)
         C = np.random.rand(m // 2)
         b = np.random.randn(m)
@@ -91,8 +91,8 @@ def time_compile_problem(self):
 class ParamSmallMatrixStuffing:
 
     def setup(self):
-        m = 100
-        n = 100
+        m = 300
+        n = 300
         A = cp.Parameter((m, n))
         C = cp.Parameter(m // 2)
         b = cp.Parameter(m)

benchmark/semidefinite_programming.py

@@ -22,8 +22,8 @@ class SemidefiniteProgramming:
 
     def setup(self):
         # Generate a random SDP.
-        n = 125
-        p = 75
+        n = 200
+        p = 120
         np.random.seed(1)
         C = np.random.randn(n, n)
         A = []

benchmark/simple_LP_benchmarks.py

@@ -31,7 +31,7 @@ def time_compile_problem(self):
 
 class SimpleFullyParametrizedLPBenchmark:
     def setup(self):
-        n = int(1e4)
+        n = int(1e6)
         p = cp.Parameter(n)
         x = cp.Variable(n)
         objective = cp.Minimize(p @ x)
@@ -45,7 +45,7 @@ def time_compile_problem(self):
 
 class SimpleScalarParametrizedLPBenchmark:
     def setup(self):
-        n = int(1e6)
+        n = int(2e6)
         p = cp.Parameter()
         c = np.arange(n)
         x = cp.Variable(n)

benchmark/simple_QP_benchmarks.py

@@ -18,9 +18,9 @@
 
 class SimpleQPBenchmark:
     def setup(self):
-        m = 2000
-        n = 400
-        p = 5
+        m = 8000
+        n = 1600
+        p = 20
         np.random.seed(1)
         P = np.random.randn(n, n)
         P = np.matmul(P.T, P)
@@ -43,8 +43,8 @@ def time_compile_problem(self):
 
 class ParametrizedQPBenchmark:
     def setup(self):
-        m = 250
-        n = 100
+        m = 6000
+        n = 2400
         np.random.seed(1)
         A = cp.Parameter((m, n))
         b = cp.Parameter((m,))
@@ -61,8 +61,8 @@ def time_compile_problem(self):
 
 class LeastSquares:
     def setup(self):
-        m = 5000
-        n = 1000
+        m = 7000
+        n = 2000
         np.random.seed(1)
         A = np.random.randn(m, n)
         b = np.random.randn(m)
@@ -81,9 +81,9 @@ class UnconstrainedQP:
     Related issue: https://github.com/cvxpy/cvxpy/issues/2205
     """
     def setup(self):
-        N_r = 16
+        N_r = 18
         N_t = 2
-        N_s = 6
+        N_s = 7
 
         x = np.random.randint(2, size=N_s * N_r * N_t)
 

benchmark/slow_pruning_1668_benchmark.py

@@ -22,10 +22,10 @@ def setup(self):
 
         Pruning matrices caused order-of-magnitude slow downs in compilation times.
         """
-        s = 2000
-        t = 10
+        s = 4000
+        t = 20
         x = np.linspace(-100.0, 100.0, s)
-        rows = 50
+        rows = 100
         var = cp.Variable(shape=(rows, t))
 
         cost = cp.sum_squares(