perf(bench): Add benchmarks for detrending, network analysis, and gravity

tests/benchmarks/test_bench_detrending.py

@@ -0,0 +1,36 @@
+"""Detrending method benchmarks: delta, identity, z-score, second-difference."""
+
+import numpy as np
+import pytest
+
+from delaynet import detrend
+
+TS_LEN = 1000
+
+_rng = np.random.default_rng(2154)
+_ts = _rng.normal(0, 1, TS_LEN)
+_ts_wave = np.sin(np.linspace(0, 4 * np.pi, TS_LEN))
+
+
+@pytest.mark.benchmark(group="detrend")
+@pytest.mark.parametrize(
+    "method,kwargs,data",
+    [
+        ("delta", {"window_size": 10}, _ts),
+        ("delta", {"window_size": 100}, _ts),
+        ("identity", {}, _ts),
+        ("z_score", {"periodicity": 1}, _ts),
+        ("z_score", {"periodicity": 10, "max_periods": 5}, _ts),
+        ("second_difference", {}, _ts_wave),
+    ],
+    ids=[
+        "delta-w10",
+        "delta-w100",
+        "identity",
+        "z_score-p1",
+        "z_score-p10",
+        "second_difference",
+    ],
+)
+def test_detrend_method(benchmark, method, kwargs, data):
+    benchmark(detrend, data, method=method, **kwargs)

tests/benchmarks/test_bench_metrics.py

@@ -5,6 +5,7 @@
 
 from delaynet.connectivities.continuous_ordinal_patterns import random_patterns
 from delaynet.connectivities.granger import gt_multi_lag
+from delaynet.connectivities.gravity import gravity
 from delaynet.connectivities.linear_correlation import linear_correlation
 from delaynet.connectivities.rank_correlation import rank_correlation
 from delaynet.connectivities.mutual_information import mutual_information
@@ -37,6 +38,11 @@ def test_continuous_metric(benchmark, metric_func, kwargs):
     benchmark(metric_func, TS1, TS2, lag_steps=LAG_STEPS, **kwargs)
 
 
+@pytest.mark.benchmark(group="metrics")
+def test_gravity(benchmark):
+    benchmark(gravity, TS1, TS2, lag_steps=LAG_STEPS, n_tests=N_TESTS)
+
+
 @pytest.mark.benchmark(group="metrics")
 def test_granger_f_test(benchmark):
     ts = gen_delayed_causal_network(ts_len=50, n_nodes=5, l_dens=0.3, rng=0)[2].T

tests/benchmarks/test_bench_network_analysis.py

@@ -0,0 +1,41 @@
+"""Network analysis metric benchmarks: global efficiency, transitivity, etc."""
+
+import numpy as np
+import pytest
+
+from delaynet.network_analysis import global_efficiency, transitivity, link_density
+
+N_NODES = 50
+DENSITY = 0.3
+
+_rng = np.random.default_rng(152)
+_adj = (_rng.uniform(0, 1, (N_NODES, N_NODES)) < DENSITY).astype(float)
+np.fill_diagonal(_adj, 0)
+
+# ensure at least some structure: a directed chain
+for i in range(N_NODES - 1):
+    _adj[i, i + 1] = 1.0
+
+# symmetric version for undirected metrics
+_adj_sym = (_adj + _adj.T) > 0
+_adj_sym = _adj_sym.astype(float)
+
+
+@pytest.mark.benchmark(group="network-analysis")
+@pytest.mark.parametrize(
+    "metric_func, kwargs, data",
+    [
+        (global_efficiency, {"directed": True}, _adj),
+        (global_efficiency, {"directed": False}, _adj_sym),
+        (transitivity, {}, _adj_sym),
+        (link_density, {"directed": True}, _adj),
+    ],
+    ids=[
+        "global_efficiency-dir",
+        "global_efficiency-undir",
+        "transitivity",
+        "link_density",
+    ],
+)
+def test_network_metric(benchmark, metric_func, kwargs, data):
+    benchmark(metric_func, data, **kwargs)