edmkit 0.0.17__tar.gz → 0.0.18__tar.gz

edmkit-0.0.18/PKG-INFO

@@ -0,0 +1,87 @@
+Metadata-Version: 2.3
+Name: edmkit
+Version: 0.0.18
+Summary: Simple EDM (Empirical Dynamic Modeling) library
+Author: FUJISHIGE TEMMA
+Author-email: FUJISHIGE TEMMA <tenma.x0@gmail.com>
+Requires-Dist: kdtree-rs>=0.3.1
+Requires-Dist: numpy>=2.5.4
+Requires-Dist: scipy>=1.18.1
+Requires-Dist: tinygrad>=0.14.0
+Requires-Dist: tinysolve>=0.3.0
+Requires-Python: >=3.13
+Description-Content-Type: text/markdown
+
+# edmkit
+
+This library is a collection of tools and utilities that are useful for Empirical Data Modeling (EDM) and related tasks. The library is designed to be fast and lightweight, and easy to use.
+
+[benchmarks](./benchmarks)
+
+![](./benchmarks/benchmarks.png)
+
+[compare.py](./benchmarks/compare.py)
+
+![](./benchmarks/compare.png)
+
+Measured on an Apple M4 Pro on November 4, 2026. Times are medians from `pytest-benchmark`. Parallel CCM uses 14 workers. Peak memory is process-tree RSS in MiB, sampled separately from timing for each case. It includes the interpreter, input, and worker processes. Summed RSS can count shared pages in more than one process.
+
+| simplex N | edmkit [ms] | pyEDM [ms] | speedup | edmkit peak [MiB] | pyEDM peak [MiB] |
+| ---: | ---: | ---: | ---: | ---: | ---: |
+| 500 | 0.17 | 1.5 | 9.0× | 168 | 169 |
+| 1,000 | 0.39 | 2.4 | 6.1× | 169 | 170 |
+| 2,000 | 0.81 | 4.4 | 5.5× | 168 | 170 |
+| 5,000 | 2.2 | 11.5 | 5.3× | 170 | 175 |
+| 10,000 | 4.7 | 23.8 | 5.0× | 172 | 180 |
+| 20,000 | 10.2 | 50.3 | 4.9× | 176 | 193 |
+
+| smap N | edmkit [ms] | pyEDM [ms] | speedup | edmkit peak [MiB] | pyEDM peak [MiB] |
+| ---: | ---: | ---: | ---: | ---: | ---: |
+| 500 | 2.1 | 36.6 | 17.9× | 176 | 180 |
+| 1,000 | 8.7 | 128.3 | 14.7× | 192 | 216 |
+| 2,000 | 22.0 | 513.6 | 23.4× | 261 | 357 |
+| 5,000 | 193.9 | 3,376.7 | 17.4× | 744 | 1,344 |
+
+| ccm N | edmkit [ms] | pyEDM [ms] | speedup | edmkit parallel [ms] | pyEDM parallel [ms] | speedup |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 500 | 106.4 | 151.7 | 1.4× | 19.5 | 68.0 | 3.5× |
+| 1,000 | 168.4 | 327.0 | 1.9× | 26.3 | 91.9 | 3.5× |
+| 2,000 | 329.7 | 745.1 | 2.3× | 48.8 | 148.3 | 3.0× |
+| 5,000 | 668.9 | 1,945.7 | 2.9× | 97.4 | 397.5 | 4.1× |
+| 10,000 | 1,253.6 | 4,176.7 | 3.3× | 164.6 | 585.4 | 3.6× |
+
+| ccm N | edmkit peak [MiB] | pyEDM peak [MiB] | edmkit parallel peak [MiB] | pyEDM parallel peak [MiB] |
+| ---: | ---: | ---: | ---: | ---: |
+| 500 | 171 | 170 | 191 | 1,417 |
+| 1,000 | 173 | 176 | 196 | 1,429 |
+| 2,000 | 179 | 186 | 207 | 1,460 |
+| 5,000 | 204 | 224 | 239 | 1,624 |
+| 10,000 | 230 | 362 | 297 | 1,804 |
+
+
+## Installation
+
+To install the library, you can use pip:
+
+```bash
+pip install edmkit
+```
+
+Or you can also use [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv add edmkit
+```
+
+## Testing
+
+```bash
+# Run all tests
+uv run pytest
+
+# Skip slow tests
+uv run pytest -m "not slow"
+
+# Run with CI profile (more hypothesis examples)
+HYPOTHESIS_PROFILE=ci uv run pytest
+```

edmkit-0.0.18/README.md

@@ -0,0 +1,73 @@
+# edmkit
+
+This library is a collection of tools and utilities that are useful for Empirical Data Modeling (EDM) and related tasks. The library is designed to be fast and lightweight, and easy to use.
+
+[benchmarks](./benchmarks)
+
+![](./benchmarks/benchmarks.png)
+
+[compare.py](./benchmarks/compare.py)
+
+![](./benchmarks/compare.png)
+
+Measured on an Apple M4 Pro on November 4, 2026. Times are medians from `pytest-benchmark`. Parallel CCM uses 14 workers. Peak memory is process-tree RSS in MiB, sampled separately from timing for each case. It includes the interpreter, input, and worker processes. Summed RSS can count shared pages in more than one process.
+
+| simplex N | edmkit [ms] | pyEDM [ms] | speedup | edmkit peak [MiB] | pyEDM peak [MiB] |
+| ---: | ---: | ---: | ---: | ---: | ---: |
+| 500 | 0.17 | 1.5 | 9.0× | 168 | 169 |
+| 1,000 | 0.39 | 2.4 | 6.1× | 169 | 170 |
+| 2,000 | 0.81 | 4.4 | 5.5× | 168 | 170 |
+| 5,000 | 2.2 | 11.5 | 5.3× | 170 | 175 |
+| 10,000 | 4.7 | 23.8 | 5.0× | 172 | 180 |
+| 20,000 | 10.2 | 50.3 | 4.9× | 176 | 193 |
+
+| smap N | edmkit [ms] | pyEDM [ms] | speedup | edmkit peak [MiB] | pyEDM peak [MiB] |
+| ---: | ---: | ---: | ---: | ---: | ---: |
+| 500 | 2.1 | 36.6 | 17.9× | 176 | 180 |
+| 1,000 | 8.7 | 128.3 | 14.7× | 192 | 216 |
+| 2,000 | 22.0 | 513.6 | 23.4× | 261 | 357 |
+| 5,000 | 193.9 | 3,376.7 | 17.4× | 744 | 1,344 |
+
+| ccm N | edmkit [ms] | pyEDM [ms] | speedup | edmkit parallel [ms] | pyEDM parallel [ms] | speedup |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 500 | 106.4 | 151.7 | 1.4× | 19.5 | 68.0 | 3.5× |
+| 1,000 | 168.4 | 327.0 | 1.9× | 26.3 | 91.9 | 3.5× |
+| 2,000 | 329.7 | 745.1 | 2.3× | 48.8 | 148.3 | 3.0× |
+| 5,000 | 668.9 | 1,945.7 | 2.9× | 97.4 | 397.5 | 4.1× |
+| 10,000 | 1,253.6 | 4,176.7 | 3.3× | 164.6 | 585.4 | 3.6× |
+
+| ccm N | edmkit peak [MiB] | pyEDM peak [MiB] | edmkit parallel peak [MiB] | pyEDM parallel peak [MiB] |
+| ---: | ---: | ---: | ---: | ---: |
+| 500 | 171 | 170 | 191 | 1,417 |
+| 1,000 | 173 | 176 | 196 | 1,429 |
+| 2,000 | 179 | 186 | 207 | 1,460 |
+| 5,000 | 204 | 224 | 239 | 1,624 |
+| 10,000 | 230 | 362 | 297 | 1,804 |
+
+
+## Installation
+
+To install the library, you can use pip:
+
+```bash
+pip install edmkit
+```
+
+Or you can also use [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv add edmkit
+```
+
+## Testing
+
+```bash
+# Run all tests
+uv run pytest
+
+# Skip slow tests
+uv run pytest -m "not slow"
+
+# Run with CI profile (more hypothesis examples)
+HYPOTHESIS_PROFILE=ci uv run pytest
+```

{edmkit-0.0.17 → edmkit-0.0.18}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "edmkit"
-version = "0.0.17"
+version = "0.0.18"
 description = "Simple EDM (Empirical Dynamic Modeling) library"
 readme = "README.md"
 requires-python = ">= 3.13"

{edmkit-0.0.17 → edmkit-0.0.18}/pyproject.toml.orig

@@ -1,6 +1,6 @@
 [project]
 name = "edmkit"
-version = "0.0.17"
+version = "0.0.18"
 description = "Simple EDM (Empirical Dynamic Modeling) library"
 authors = [{ name = "FUJISHIGE TEMMA", email = "tenma.x0@gmail.com" }]
 readme = "README.md"

{edmkit-0.0.17 → edmkit-0.0.18}/src/edmkit/simplex_projection/knn.py

@@ -32,38 +32,34 @@ def knn(X: np.ndarray, Q: np.ndarray, k: int, mask: Mask | None = None) -> tuple
     if N < k:
         raise ValueError(f"Not enough points in X to find {k} neighbors, got N={N}")
 
-    tree = KDTree(X)
-    if mask is None:
+    # smaller leaves improve pruning in low dimensions.
+    tree = KDTree(X, leafsize=16 if X.shape[-1] <= 8 else 32)
+    if mask is None or len(Q) == 0:
         return tree.query(Q, k=k)
 
     distances = np.empty((Q.shape[0], k))
     indices = np.empty((Q.shape[0], k), dtype=np.intp)
 
-    def fetch(pending: np.ndarray, k_query: int) -> np.ndarray:
-        """Fill `distances` and `indices` with the k nearest neighbors for the `pending` queries, returning the queries that still need more neighbors."""
+    pending = np.arange(Q.shape[0])
+    k_query = min(k + 1, N)
 
-        # over fetch (k_query > k)
-        _distances, _indices = tree.query(Q[pending], k=k_query)
+    while pending.size > 0:
+        _distances, _indices = tree.query(Q if pending.size == len(Q) else Q[pending], k=k_query)
 
-        # evaluate the mask and filter the resultss
+        # evaluate the mask and keep the first k valid candidates.
         valid = np.broadcast_to(np.asarray(mask(pending[:, None], _indices), dtype=bool), _indices.shape)
         count = valid.sum(axis=-1)
 
         if k_query == N and count.min() < k:
             raise ValueError(f"Not enough unmasked points in X to find {k} neighbors, got {count.min()} for some query")
 
-        # take the first k valid neighbors for the queries that have them
         done = count >= k
-        first_k = valid[done] & (np.cumsum(valid[done], axis=-1) <= k)
-        distances[pending[done]] = _distances[done][first_k].reshape(-1, k)
-        indices[pending[done]] = _indices[done][first_k].reshape(-1, k)
-
-        return pending[~done]
+        if done.any():
+            first_k = valid & (valid.cumsum(axis=-1, dtype=np.uint32) <= k) & done[:, None]
+            distances[pending[done]] = _distances[first_k].reshape(-1, k)
+            indices[pending[done]] = _indices[first_k].reshape(-1, k)
 
-    k_query = min(k + 1, N)
-    pending = fetch(np.arange(Q.shape[0]), k_query)
-    while pending.size > 0:
+        pending = pending[~done]
         k_query = min(2 * k_query, N)
-        pending = fetch(pending, k_query)
 
     return distances, indices

{edmkit-0.0.17 → edmkit-0.0.18}/src/edmkit/simplex_projection/simplex_projection.py

@@ -145,8 +145,6 @@ def _numpy(
     # treat the unbatched case as a single batch
     if X.ndim == 2:
         X, Y, Q = X[None], Y[None], Q[None]
-        if mask is not None:
-            mask = lambda _, *index, mask=mask: mask(*index)
 
     B, _, E = X.shape
     M = Q.shape[1]
@@ -154,19 +152,18 @@ def _numpy(
     if k <= 0:
         raise ValueError(f"k must be positive, got k={k}")
 
-    distances = np.empty((B, M, k))
-    indices = np.empty((B, M, k), dtype=np.intp)
+    predictions = np.empty((B, M, Y.shape[-1]), dtype=np.result_type(np.float64, Y.dtype))
     for b in range(B):
-        distances[b], indices[b] = knn(X[b], Q[b], k, None if mask is None else partial(mask, b))
+        distances, indices = knn(X[b], Q[b], k, partial(mask, b) if return_ndim == 3 and mask is not None else mask)
 
-    batch_index = np.arange(B)[:, None, None]  # (B, 1, 1)
-    Y_neighbors = Y[batch_index, indices]  # (B, M, k, E')
+        # Neighbors are sorted. Clamp the scale and reuse the distance buffer.
+        d_min = np.fmax(distances[:, :1], 1e-6)
+        np.divide(distances, -d_min, out=distances)
+        weights = np.exp(distances, out=distances)
 
-    # clamp to avoid division by zero
-    d_min = np.fmax(distances.min(axis=-1, keepdims=True), 1e-6)  # (B, M, 1)
-    weights = np.exp(-distances / d_min)  # (B, M, k)
-
-    predictions = np.matmul(weights[..., None, :], Y_neighbors).squeeze(-2) / weights.sum(axis=-1, keepdims=True)  # (B, M, E')
+        prediction = predictions[b]
+        np.matmul(weights[:, None, :], Y[b, indices], out=prediction[:, None, :])
+        prediction /= weights.sum(axis=-1, keepdims=True)
 
     if return_ndim == 1:
         return predictions[0, :, 0]  # (M,)

edmkit-0.0.17/PKG-INFO

@@ -1,77 +0,0 @@
-Metadata-Version: 2.3
-Name: edmkit
-Version: 0.0.17
-Summary: Simple EDM (Empirical Dynamic Modeling) library
-Author: FUJISHIGE TEMMA
-Author-email: FUJISHIGE TEMMA <tenma.x0@gmail.com>
-Requires-Dist: kdtree-rs>=0.3.1
-Requires-Dist: numpy>=2.5.4
-Requires-Dist: scipy>=1.18.1
-Requires-Dist: tinygrad>=0.14.0
-Requires-Dist: tinysolve>=0.3.0
-Requires-Python: >=3.13
-Description-Content-Type: text/markdown
-
-# edmkit
-
-This library is a collection of tools and utilities that are useful for Empirical Data Modeling (EDM) and related tasks. The library is designed to be fast and lightweight, and easy to use.
-
-[benchmarks](./benchmarks)
-
-![](./benchmarks/benchmarks.png)
-
-[compare.py](./benchmarks/compare.py)
-
-![](./benchmarks/compare.png)
-
-| simplex N | edmkit [ms] | pyEDM [ms] | speedup |
-|---:|---:|---:|---:|
-| 500 | 0.2 | 1.4 | 8x |
-| 1,000 | 0.4 | 2.3 | 5x |
-| 2,000 | 0.9 | 4.2 | 5x |
-| 5,000 | 2.3 | 10.7 | 5x |
-| 10,000 | 4.9 | 22.8 | 5x |
-| 20,000 | 10.3 | 46.8 | 5x |
-
-| smap N | edmkit [ms] | pyEDM [ms] | speedup |
-|---:|---:|---:|---:|
-| 500 | 1.9 | 38.0 | 20x |
-| 1,000 | 7.5 | 132.1 | 18x |
-| 2,000 | 21.6 | 522.6 | 24x |
-| 5,000 | 183.7 | 3,364 | 18x |
-
-| ccm N | edmkit [ms] | pyEDM [ms] | speedup | edmkit parallel [ms] | pyEDM parallel [ms] | speedup |
-|---:|---:|---:|---:|---:|---:|---:|
-| 500 | 116.7 | 148.4 | 1.3x | 22.0 | 64.5 | 2.9x |
-| 1,000 | 197.1 | 316.8 | 1.6x | 31.8 | 89.7 | 2.8x |
-| 2,000 | 382.8 | 696.8 | 1.8x | 58.6 | 133.8 | 2.3x |
-| 5,000 | 788.0 | 1,863 | 2.4x | 108.9 | 283.5 | 2.6x |
-| 10,000 | 1,470 | 4,166 | 2.8x | 197.0 | 558.9 | 2.8x |
-
-
-## Installation
-
-To install the library, you can use pip:
-
-```bash
-pip install edmkit
-```
-
-Or you can also use [uv](https://docs.astral.sh/uv/):
-
-```bash
-uv add edmkit
-```
-
-## Testing
-
-```bash
-# Run all tests
-uv run pytest
-
-# Skip slow tests
-uv run pytest -m "not slow"
-
-# Run with CI profile (more hypothesis examples)
-HYPOTHESIS_PROFILE=ci uv run pytest
-```

edmkit-0.0.17/README.md

@@ -1,63 +0,0 @@
-# edmkit
-
-This library is a collection of tools and utilities that are useful for Empirical Data Modeling (EDM) and related tasks. The library is designed to be fast and lightweight, and easy to use.
-
-[benchmarks](./benchmarks)
-
-![](./benchmarks/benchmarks.png)
-
-[compare.py](./benchmarks/compare.py)
-
-![](./benchmarks/compare.png)
-
-| simplex N | edmkit [ms] | pyEDM [ms] | speedup |
-|---:|---:|---:|---:|
-| 500 | 0.2 | 1.4 | 8x |
-| 1,000 | 0.4 | 2.3 | 5x |
-| 2,000 | 0.9 | 4.2 | 5x |
-| 5,000 | 2.3 | 10.7 | 5x |
-| 10,000 | 4.9 | 22.8 | 5x |
-| 20,000 | 10.3 | 46.8 | 5x |
-
-| smap N | edmkit [ms] | pyEDM [ms] | speedup |
-|---:|---:|---:|---:|
-| 500 | 1.9 | 38.0 | 20x |
-| 1,000 | 7.5 | 132.1 | 18x |
-| 2,000 | 21.6 | 522.6 | 24x |
-| 5,000 | 183.7 | 3,364 | 18x |
-
-| ccm N | edmkit [ms] | pyEDM [ms] | speedup | edmkit parallel [ms] | pyEDM parallel [ms] | speedup |
-|---:|---:|---:|---:|---:|---:|---:|
-| 500 | 116.7 | 148.4 | 1.3x | 22.0 | 64.5 | 2.9x |
-| 1,000 | 197.1 | 316.8 | 1.6x | 31.8 | 89.7 | 2.8x |
-| 2,000 | 382.8 | 696.8 | 1.8x | 58.6 | 133.8 | 2.3x |
-| 5,000 | 788.0 | 1,863 | 2.4x | 108.9 | 283.5 | 2.6x |
-| 10,000 | 1,470 | 4,166 | 2.8x | 197.0 | 558.9 | 2.8x |
-
-
-## Installation
-
-To install the library, you can use pip:
-
-```bash
-pip install edmkit
-```
-
-Or you can also use [uv](https://docs.astral.sh/uv/):
-
-```bash
-uv add edmkit
-```
-
-## Testing
-
-```bash
-# Run all tests
-uv run pytest
-
-# Skip slow tests
-uv run pytest -m "not slow"
-
-# Run with CI profile (more hypothesis examples)
-HYPOTHESIS_PROFILE=ci uv run pytest
-```