initial commit: ts time series analysis library v0.1.0

- 354 financial and mathematical time-series indicators
- Uniform C11 API with consistent function signatures
- 349 passing tests with -Werror
- Static and shared library builds
- pkg-config integration
- Python bindings
- Fuzzing and sanitizer support
- Security hardening flags

1 files changed, 122 insertions, 0 deletions

diff --git a/lib/hurst_exponent.c b/lib/hurst_exponent.c
new file mode 100644
index 0000000..aa58036
--- /dev/null
+++ b/lib/hurst_exponent.c
@@ -0,0 +1,122 @@
+// Hurst Exponent
+//
+// hurst_exponent
+//
+// Hurst exponent via R/S analysis
+//
+// H = slope of linear regression of log(R/S) vs log(n)
+//
+// Inputs:    1
+// Outputs:   1
+// Constants: 1
+
+
+#include <math.h>
+#include "ts.h"
+
+typedef struct {
+    uint_t k;
+} ts_hurst_exponent_params_t;
+
+ts_parameter_t const ts_hurst_exponent_p = { .n_inputs = 1, .n_outputs = 1, .n_constants = 1, .params_size = sizeof(ts_hurst_exponent_params_t) };
+
+uint_t ts_hurst_exponent_offset( ts_constant_t const *c )
+{
+	( void )c;
+	return ( 0 );
+}
+
+int ts_hurst_exponent(
+	fp_t * const *y, uint_t y_n,
+	fp_t const * const *x, uint_t x_n,
+	uint_t n, ts_constant_t const *c
+)
+{
+	TS_PROLOGUE( ts_hurst_exponent_p );
+	if ( !c ) return -1;
+    uint_t k = ((ts_hurst_exponent_params_t *)c)->k;
+    if (k == 0 || k > n) return -1;
+
+    for ( uint_t i = 0; i < n; ++i )
+    {
+        if (i + 1 < k) {
+            y[0][i] = 0;
+            continue;
+        }
+
+        uint_t data_len = i + 1;
+        uint_t num_lags = 0;
+        fp_t sum_log_s = 0;
+        fp_t sum_log_rs = 0;
+        fp_t sum_log_s2 = 0;
+        fp_t sum_log_s_log_rs = 0;
+
+        for ( uint_t s = 2; s <= k && s <= data_len / 2; ++s )
+        {
+            uint_t ns = data_len / s;
+            if (ns == 0) continue;
+
+            fp_t total_log_rs = 0;
+            uint_t valid_segs = 0;
+
+            for ( uint_t seg = 0; seg < ns; ++seg )
+            {
+                uint_t start = seg * s;
+                uint_t end = start + s;
+                if (end > data_len) break;
+
+                fp_t mean = 0;
+                for ( uint_t p = start; p < end; ++p )
+                    mean += x[0][p];
+                mean /= (fp_t)s;
+
+                fp_t cumsum = 0;
+                fp_t max_dev = 0, min_dev = 0;
+                for ( uint_t p = start; p < end; ++p )
+                {
+                    cumsum += x[0][p] - mean;
+                    if (cumsum > max_dev) max_dev = cumsum;
+                    if (cumsum < min_dev) min_dev = cumsum;
+                }
+
+                fp_t r = max_dev - min_dev;
+                fp_t std = 0;
+                for ( uint_t p = start; p < end; ++p )
+                {
+                    fp_t d = x[0][p] - mean;
+                    std += d * d;
+                }
+                std = sqrt(std / (fp_t)s);
+
+                if (std > 0 && r > 0) {
+                    total_log_rs += log(r / std);
+                    valid_segs++;
+                }
+            }
+
+            if (valid_segs > 0) {
+                fp_t log_s = log((fp_t)s);
+                fp_t mean_log_rs = total_log_rs / (fp_t)valid_segs;
+                sum_log_s += log_s;
+                sum_log_rs += mean_log_rs;
+                sum_log_s2 += log_s * log_s;
+                sum_log_s_log_rs += log_s * mean_log_rs;
+                num_lags++;
+            }
+        }
+
+        if (num_lags >= 2) {
+            fp_t denom = (fp_t)num_lags * sum_log_s2 - sum_log_s * sum_log_s;
+            if (fabs(denom) > 1e-300) {
+                fp_t h = ((fp_t)num_lags * sum_log_s_log_rs - sum_log_s * sum_log_rs) / denom;
+                y[0][i] = h;
+            } else {
+                y[0][i] = 0;
+            }
+        } else {
+            y[0][i] = 0;
+        }
+    }
+	return 0;
+}
+