Use pixi to install environment

.github/workflows/pixi_test.yaml

@@ -0,0 +1,31 @@
+name: Tests on GPU
+
+on:
+  push:
+    branches:
+      - main
+    tags:
+      - v*.*.*
+  pull_request: {}
+
+jobs:
+  test:
+    runs-on: [ubuntu-latest, gpu]
+
+    steps:
+        - name: Check out the repository
+          uses: actions/checkout@v4
+
+        - name: Install pixi
+          run: |
+            curl -fsSL https://pixi.sh/install.sh | bash
+            echo "$HOME/.pixi/bin" >> $GITHUB_PATH
+
+        - name: Show NVIDIA-smi
+          run: nvidia-smi
+
+        - name: install pixi test environment
+          run: pixi install -e dev
+
+        - name: run pytest
+          run: pixi run -e dev test

bigwig_loader/intervals_to_values.py

@@ -1,5 +1,6 @@
 import logging
 import math
+from math import isnan
 from pathlib import Path
 
 import cupy as cp
@@ -120,6 +121,7 @@ def intervals_to_values(
     array_start = cp.ascontiguousarray(array_start)
     array_end = cp.ascontiguousarray(array_end)
     array_value = cp.ascontiguousarray(array_value)
+    default_value_isnan = isnan(default_value)
 
     cuda_kernel(
         (grid_size,),
@@ -137,6 +139,8 @@ def intervals_to_values(
             sequence_length,
             max_number_intervals,
             window_size,
+            default_value,
+            default_value_isnan,
             out,
         ),
     )

cuda_kernels/intervals_to_values.cu

@@ -1,3 +1,5 @@
+#include <math_constants.h>
+
 extern "C" __global__
 void intervals_to_values(
         const unsigned int* query_starts,
@@ -12,6 +14,8 @@ void intervals_to_values(
         const int sequence_length,
         const int max_number_intervals,
         const int window_size,
+        const float default_value,
+        const bool default_value_isnan,
         float* out
 ) {
 
@@ -49,7 +53,7 @@ void intervals_to_values(
 		}
 	} else {
 
-		int track_index = i / batch_size;
+// 		int track_index = i / batch_size;
 
 		int found_start_index = found_starts[i];
 		int found_end_index = found_ends[i];
@@ -59,6 +63,8 @@ void intervals_to_values(
 		int cursor = found_start_index;
 		int window_index = 0;
 		float summation = 0.0f;
+		int valid_count = 0;
+
 
 		int reduced_dim = sequence_length / window_size;
 
@@ -73,19 +79,34 @@ void intervals_to_values(
 			int end_index = min(interval_end, query_end) - query_start;
 
 			if (start_index >= window_end) {
-				out[i * reduced_dim + window_index] = summation / window_size;
+			    if (default_value_isnan) {
+					out[i * reduced_dim + window_index] = valid_count > 0 ? summation / valid_count : CUDART_NAN_F;
+			    } else {
+			        summation = summation + (window_size - valid_count) * default_value;
+			        out[i * reduced_dim + window_index] = summation / window_size;
+			    }
 				summation = 0.0f;
+				valid_count = 0;
 				window_index += 1;
 				continue;
 			}
 
 			int number = min(window_end, end_index) - max(window_start, start_index);
 
-			summation += number * track_values[cursor];
+			if (number > 0) {
+				summation += number * track_values[cursor];
+				valid_count += number;
+			}
 
 			if (end_index >= window_end || cursor + 1 >= found_end_index) {
-				out[i * reduced_dim + window_index] = summation / window_size;
-				summation = 0.0f;
+			    if (default_value_isnan) {
+					out[i * reduced_dim + window_index] = valid_count > 0 ? summation / valid_count : CUDART_NAN_F;
+			    } else {
+			        summation = summation + (window_size - valid_count) * default_value;
+			        out[i * reduced_dim + window_index] = summation / window_size;
+			    }
+			    summation = 0.0f;
+				valid_count = 0;
 				window_index += 1;
 			}
 

example_data/some_positions.tsv

@@ -1,4 +1,5 @@
 chr	center
+chr1	45298878
 chr18	61036865
 chr17	12174372
 chr3	65857025

pixi.toml

@@ -0,0 +1,44 @@
+[workspace]
+channels = ["rapidsai", "conda-forge", "nvidia"]
+name = "bigwig-loader"
+platforms = ["linux-64"]
+version = "0.1.0"
+
+[tasks]
+test = { cmd = 'pytest -v' }
+
+[dependencies]
+python = ">=3.10"
+pip = "*"
+cython="*"
+c-compiler = "*"
+kvikio = ">=24.06.00"
+cupy = ">=12.0.0"
+numpy = ">=1.21"
+pandas = "*"
+
+[pypi-dependencies]
+torch = "*"
+python-dotenv = "*"
+pydantic = "*"
+pydantic-settings = "*"
+pyfaidx = "*"
+natsort = "*"
+setuptools-scm = "*"
+ncls = "*"
+bigwig_loader = { path = ".", editable = true }
+
+[feature.test.pypi-dependencies]
+pytest = "*"
+pytest-sugar = "*"
+pytest-cov = "*"
+pytest-mock = "*"
+pyBigWig = "*"
+
+[feature.dev.pypi-dependencies]
+mypy = "*"
+pre-commit = "*"
+
+[environments]
+test = {features = ["test"]}
+dev = {features = ["test", "dev"]}

tests/test_against_pybigwig.py

@@ -26,8 +26,8 @@ def get_batch(self, chromosomes, starts, ends):
 
 
 def test_same_output(bigwig_path):
-    pybigwig_collection = PyBigWigCollection(bigwig_path, first_n_files=2)
-    collection = BigWigCollection(bigwig_path, first_n_files=2)
+    pybigwig_collection = PyBigWigCollection(bigwig_path, first_n_files=3)
+    collection = BigWigCollection(bigwig_path, first_n_files=3)
 
     df = pd.read_csv(config.example_positions, sep="\t")
     df = df[df["chr"].isin(collection.get_chromosomes_present_in_all_files())]
@@ -49,3 +49,31 @@ def test_same_output(bigwig_path):
     print(this_batch[pybigwig_batch != this_batch])
     print(pybigwig_batch[pybigwig_batch != this_batch])
     assert (pybigwig_batch == this_batch).all()
+
+
+def test_same_output_with_nans(bigwig_path):
+    pybigwig_collection = PyBigWigCollection(bigwig_path, first_n_files=3)
+    collection = BigWigCollection(bigwig_path, first_n_files=3)
+
+    df = pd.read_csv(config.example_positions, sep="\t")
+    df = df[df["chr"].isin(collection.get_chromosomes_present_in_all_files())]
+    chromosomes, starts, ends = (
+        list(df["chr"]),
+        list(df["center"] - 1000),
+        list(df["center"] + 1000),
+    )
+
+    pybigwig_batch = pybigwig_collection.get_batch(chromosomes, starts, ends)
+
+    this_batch = collection.get_batch(
+        chromosomes, starts, ends, default_value=np.nan
+    ).get()
+    print("PyBigWig:")
+    print(pybigwig_batch)
+    print(type(this_batch), "shape:", pybigwig_batch.shape)
+    print("This Library:")
+    print(this_batch)
+    print(type(this_batch), "shape:", this_batch.shape)
+    print(this_batch[pybigwig_batch != this_batch])
+    print(pybigwig_batch[pybigwig_batch != this_batch])
+    assert np.allclose(pybigwig_batch, this_batch, equal_nan=True)