Examples, stats, and tests

convergentist · convergentist · commit af12b9ccb898 · 2026-03-20T00:58:27.000-06:00
diff --git a/atelier-data/Cargo.toml b/atelier-data/Cargo.toml
@@ -23,7 +23,7 @@ publish = true
 
 [package.metadata.docs.rs]
 rustdoc-args = ["--html-in-header", "katex-header.html", "--cfg", "docsrs"]
-all-features = true
+features = ["parquet"]
 
 [lib]
 name = "atelier_data"
diff --git a/atelier-data/examples/market_worker/market_worker_binance.toml b/atelier-data/examples/market_worker/market_worker_binance.toml
@@ -44,8 +44,5 @@ type = "terminal"
 
 # ── Workers ──────────────────────────────────────────────────────────
 
-[[workers]]
-symbol = "ethusdt"
-
 [[workers]]
 symbol = "solusdt"
diff --git a/atelier-data/examples/market_worker/market_worker_coinbase.toml b/atelier-data/examples/market_worker/market_worker_coinbase.toml
@@ -44,8 +44,5 @@ type = "terminal"
 
 # ── Workers ─────────────────────────────────────────────────────────────
 
-[[workers]]
-symbol = "ETH-USDT"
-
 [[workers]]
 symbol = "SOL-USDT"
diff --git a/atelier-data/examples/market_worker/market_worker_kraken.toml b/atelier-data/examples/market_worker/market_worker_kraken.toml
@@ -44,8 +44,5 @@ type = "terminal"
 
 # ── Workers ─────────────────────────────────────────────────────────────
 
-[[workers]]
-symbol = "ETH/USDT"
-
 [[workers]]
 symbol = "SOL/USDT"
diff --git a/atelier-quant/artifacts/inter_arrival/experiments/hawkes_fit_20260312_060842.json b/atelier-quant/artifacts/inter_arrival/experiments/hawkes_fit_20260312_060842.json
@@ -0,0 +1,35 @@
+{
+  "model": "hawkes_exponential",
+  "version": 1,
+  "fitted_at": "2026-03-12T06:08:42.919498Z",
+  "parameters": {
+    "mu": 6.7782254489266944e-6,
+    "alpha": 6.144956410640459e-6,
+    "beta": 0.000010190009253087155
+  },
+  "diagnostics": {
+    "branching_ratio": 0.6030373729816574,
+    "log_likelihood": -669.5290716452546,
+    "aic": 1345.0581432905092,
+    "bic": 1351.1341983627146,
+    "converged": true,
+    "iterations": 64,
+    "residuals_mean": 1.018315294732614,
+    "residuals_std": 1.2302185316918353,
+    "compensator_ratio": 1.018315294732614
+  },
+  "data": {
+    "source": "datasets/collected/coinbase/trades/coinbase_SOL-USDT_trades_sync_20260306_181353.735.parquet",
+    "n_events_train": 56,
+    "n_events_test": 2,
+    "observation_window_ms": 3525335.0,
+    "time_resolution": "Milliseconds"
+  },
+  "baseline": {
+    "model": "poisson",
+    "lambda": 0.000015847303451657945,
+    "aic": 1327.7762321482294,
+    "lr_statistic": -13.281911142279796,
+    "lr_reject_h0": false
+  }
+}
diff --git a/atelier-quant/artifacts/inter_arrival/experiments/hawkes_fit_20260312_060948.json b/atelier-quant/artifacts/inter_arrival/experiments/hawkes_fit_20260312_060948.json
@@ -0,0 +1,35 @@
+{
+  "model": "hawkes_exponential",
+  "version": 1,
+  "fitted_at": "2026-03-12T06:09:48.698845Z",
+  "parameters": {
+    "mu": 0.000016077261288393247,
+    "alpha": 0.0003266350897340653,
+    "beta": 0.0012323393472207275
+  },
+  "diagnostics": {
+    "branching_ratio": 0.2650528772539151,
+    "log_likelihood": -849.1563448323541,
+    "aic": 1704.3126896647082,
+    "bic": 1711.265154005317,
+    "converged": true,
+    "iterations": 34987,
+    "residuals_mean": 1.0123360363238076,
+    "residuals_std": 2.33933084957549,
+    "compensator_ratio": 1.0123360363238076
+  },
+  "data": {
+    "source": "datasets/collected/kraken/trades/kraken_SOL-USDT_trades_sync_20260306_181411.334.parquet",
+    "n_events_train": 75,
+    "n_events_test": 2,
+    "observation_window_ms": 3460600.0,
+    "time_resolution": "Milliseconds"
+  },
+  "baseline": {
+    "model": "poisson",
+    "lambda": 0.000021428170167469836,
+    "aic": 1741.1190125152762,
+    "lr_statistic": 40.806322850567994,
+    "lr_reject_h0": true
+  }
+}
diff --git a/atelier-quant/configs/inter_fit.toml b/atelier-quant/configs/inter_fit.toml
@@ -11,7 +11,7 @@
 
 [input]
 # Directory containing parquet files.
-path = "datasets/collected/bybit/trades/"
+path = "datasets/collected/kraken/trades/"
 # Data type to look for: "trades" or "orderbook".
 data_type = "trades"
 # File selection: "latest" picks the most recently modified file.
@@ -20,13 +20,13 @@ selection = "latest"
 [model]
 # MLE optimizer settings.
 max_iter = 50_000
-tolerance = 1e2
-learning_rate = 2e-3
+tolerance = 1e3
+learning_rate = 1e-3
 # Fraction of events used for training (remainder is the test set).
-train_ratio = 0.986
+train_ratio = 0.980
 # Gap detection threshold (seconds). Gaps above this are flagged
 # as likely feed disconnects.
-gap_threshold_secs = 5.0
+gap_threshold_secs = 9.0
 
 [forecast]
 # Number of Monte-Carlo stochastic paths (1 = single run, >1 = ensemble).
@@ -36,7 +36,7 @@ mc_statistic = "p25"
 
 [output]
 # Directory where the model artifact JSON will be written.
-artifact_dir = "artifacts/inter_arrival/"
+artifact_dir = "atelier-quant/artifacts/inter_arrival/experiments/"
 # Output format (only "json" is currently supported).
 format = "json"
 # Include goodness-of-fit diagnostics in the artifact.
diff --git a/atelier-quant/src/arrivals/inter.rs b/atelier-quant/src/arrivals/inter.rs
@@ -203,4 +203,3 @@ pub fn descriptive_stats(deltas: &[f64]) -> Option<ArrivalStats> {
         cv,
     })
 }
-
diff --git a/atelier-quant/src/forecast.rs b/atelier-quant/src/forecast.rs
@@ -8,8 +8,8 @@
 //! - [`percentile`] — linearly interpolated quantile from a sorted slice.
 
 use crate::config::EnsembleStatistic;
-use crate::hawkes::kernel::ExcitationKernel;
 use crate::hawkes::HawkesProcess;
+use crate::hawkes::kernel::ExcitationKernel;
 use serde::{Deserialize, Serialize};
 
 /// Forecast error metrics computed on cumulative gaps from the
@@ -110,7 +110,9 @@ pub fn ensemble_forecast<K: ExcitationKernel>(
                     }
                 }
                 _ => {
-                    column.sort_unstable_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
+                    column.sort_unstable_by(|a, b| {
+                        a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal)
+                    });
                     let q = match statistic {
                         EnsembleStatistic::Median => 0.5,
                         EnsembleStatistic::P25 => 0.25,
@@ -172,7 +174,11 @@ pub struct LRTestResult {
 /// # Returns
 ///
 /// [`LRTestResult`] with the test statistic and decision at α = 0.05.
-pub fn likelihood_ratio_test(ll_full: f64, ll_restricted: f64, df: usize) -> LRTestResult {
+pub fn likelihood_ratio_test(
+    ll_full: f64,
+    ll_restricted: f64,
+    df: usize,
+) -> LRTestResult {
     let statistic = 2.0 * (ll_full - ll_restricted);
 
     // χ²(df) critical values at α = 0.05 for small df.
diff --git a/atelier-quant/src/hawkes/mod.rs b/atelier-quant/src/hawkes/mod.rs
@@ -14,9 +14,9 @@
 //! - [`errors`] — error types for invalid parameters and convergence
 //!   failures.
 
+pub mod core;
 pub mod kernel;
 pub mod quadrature;
-pub mod core;
 pub use core::HawkesProcess;
 pub mod errors;
 pub mod estimation;
diff --git a/atelier-quant/src/hawkes/quadrature.rs b/atelier-quant/src/hawkes/quadrature.rs
@@ -131,4 +131,3 @@ pub fn conditional_mean_gap<F: Fn(f64) -> f64>(
     // The (1/μ) factor comes from the ds = dx/μ substitution.
     inv_mu * adaptive_simpson(&integrand, 0.0, x_max, tol, 20)
 }
-
diff --git a/atelier-quant/src/poisson/estimation.rs b/atelier-quant/src/poisson/estimation.rs
@@ -84,7 +84,9 @@ pub fn validate_events(events: &[f64]) -> Result<(), PoissonError> {
         if events[i] <= events[i - 1] {
             return Err(PoissonError::InvalidEventTimes(format!(
                 "Events not strictly increasing at index {}: {} <= {}",
-                i, events[i], events[i - 1]
+                i,
+                events[i],
+                events[i - 1]
             )));
         }
     }
diff --git a/atelier-quant/src/stats/eda.rs b/atelier-quant/src/stats/eda.rs
@@ -0,0 +1,34 @@
+
+pub us std::collections::HashMap;
+
+
+/// 4.1
+///
+/// Number of Trade-Through Market Orders
+
+
+/// Table 4.1 Summary Statistics of Aggregate Activity, (Bouchaud et all, 2018)
+///
+/// Average share price
+/// Average quoted spread
+/// Average spread before transactions
+/// Fraction of MO that match at worse prices than the best quote
+/// Fraction of MO that match a hidden order inside the spread
+/// Mean total volume at best quotes (dollars x 1e3)
+/// Mean daily traded volume on NASDAQ†(dollars x 1e6)
+/// Mean fraction of daily traded volume that is hidden
+/// Mean total volume of active orders within 1% of mid-price (dollars x 1e6)
+/// Average stock-specific market share of NASDAQ†
+/// Average total market capitalisation†(dollars x 1e9)
+/// Average daily number (Best Quote Orders x 1e3) (LO, MO, Cancels)
+/// Mean Inter-Arrival Time of Best Quote Orders (seconds)
+/// Median Inter-Arrival Time of Best Quote Orders (seconds)
+/// Mean Size of Best Quote Orders (dollars x 1e3)
+///
+
+
+/// 4.2
+///
+/// Intraday patterns
+///
+///
diff --git a/atelier-quant/src/stats/gaps.rs b/atelier-quant/src/stats/gaps.rs
@@ -0,0 +1,37 @@
+
+use std::collections::HashMap;
+// Detect gaps > 5 seconds (5_000_000_000 ns) — likely feed disconnects
+let gap_threshold_ns = 5_000_000_000_u64;
+
+
+/// Returns the list of indexes where the timestamp differences was 
+/// greater than a threshold parameter expressed in nanoseconds
+pub fn list_gaps(gap_threshold_ns: u64) -> Result<HashMap, err::Err>{
+
+    let mut n_gaps = 0_usize;
+    let mapped_gaps = HashMap::new();
+
+    for i in 1..timestamps_ns.len() {
+
+        let gap = timestamps_ns[i] - timestamps_ns[i - 1];
+
+        if gap > gap_threshold_ns {
+
+            mapped_gaps.append(i, gap);
+
+            println!("    index {}: gap = {:.3} ms", i - 1, gap as f64 / 1e6);
+            n_gaps += 1;
+        }
+    }
+
+    if n_gaps == 0 {
+        println!(
+            "  No gaps exceeding {:.1}s detected", gap_threshold_ns as f64 / 1e9
+        );
+    } else {
+        println!("  Total large gaps: {}", n_gaps);
+    }
+
+}
+
+
diff --git a/atelier-quant/src/stats/mod.rs b/atelier-quant/src/stats/mod.rs
@@ -0,0 +1,7 @@
+//! # Stats 
+//!
+//! ## Gap Detection
+//!
+
+pub mod gaps;
+
diff --git a/atelier-quant/src/stats/orders.rs b/atelier-quant/src/stats/orders.rs
@@ -0,0 +1,17 @@
+/// Market Orders Stream Reconstruction 
+/// from OB and PT
+/// a PT at t is a Market Order executed on the OB at t-1
+
+
+/// Limit Orders Stream Reconstruction
+/// from DOB
+/// each level's volume represents the arithmetic sum of New LO and Cancel LO
+/// volume is a summation, count is unknown
+///
+/// How to derive the count of orders per level, from L2 data, is an open question.
+///
+/// Approaches: 
+///     Naive: At maximum, No. of orders = level's volume / min quotable volume
+///     TTO: trade flow as a proxy to order count  
+///     OTO: order flow in DOB to order count
+
diff --git a/atelier-quant/tests/hawkes/test_hawkes.rs b/atelier-quant/tests/hawkes/test_hawkes.rs
@@ -1,6 +1,6 @@
 #[cfg(test)]
 mod tests {
-    use atelier_quant::{hawkes::errors::HawkesError, hawkes::HawkesProcess};
+    use atelier_quant::{hawkes::HawkesProcess, hawkes::errors::HawkesError};
 
     /// Test 1: HawkesInputTypeFailure
     /// Tests invalid input parameters that should fail validation
diff --git a/atelier-quant/tests/poisson/test_estimation.rs b/atelier-quant/tests/poisson/test_estimation.rs
@@ -1,7 +1,7 @@
 #[cfg(test)]
 mod tests {
-    use atelier_quant::poisson::estimation;
     use atelier_quant::poisson::PoissonProcess;
+    use atelier_quant::poisson::estimation;
 
     // ── validate_events ──────────────────────────────────────────────
 
@@ -39,11 +39,7 @@ mod tests {
     fn test_log_likelihood_known_value() {
         let events = vec![0.0, 1.0, 2.0, 3.0];
         let ll = estimation::log_likelihood(1.0, &events);
-        assert!(
-            (ll - (-3.0)).abs() < 1e-12,
-            "Expected -3.0, got {}",
-            ll
-        );
+        assert!((ll - (-3.0)).abs() < 1e-12, "Expected -3.0, got {}", ll);
     }
 
     #[test]
@@ -225,11 +221,7 @@ mod tests {
 
         assert_eq!(residuals.len(), 3);
         for &r in &residuals {
-            assert!(
-                (r - 2.0).abs() < 1e-12,
-                "Expected residual 2.0, got {}",
-                r
-            );
+            assert!((r - 2.0).abs() < 1e-12, "Expected residual 2.0, got {}", r);
         }
     }
 }

Original file line number	Diff line number	Diff line change
`@@ -203,4 +203,3 @@ pub fn descriptive_stats(deltas: &[f64]) -> Option<ArrivalStats> {`
`203`	`203`	`cv,`
`204`	`204`	`})`
`205`	`205`	`}`
`206`		`-`