added basic tests & updated readmes

README.md

@@ -63,18 +63,7 @@ To install the DRP along with its dependencies, you need to run the following st
     cd lvmdrp
     ```
 
-3. Optional, but recommended: compile c-code for better performance
-
-    ```bash
-    pushd lvmdrp/python/cextern/fast_median/src
-    make
-    popd
-    ```
-    This should have created a fast_median.so (Linux) or fast_median.dylib (MacOS) in the src directory.
-    Set environment variable `LVMDRP_LIB_DIR` if you move the shared library. If you leave it in src/, no need to set this.
-    We will automate this step eventually ;-)
-
-4. Install the DRP package in the current python environment (see [contributing](#contributing-to-lvm-drp-development) section below for a replacement of this step). Make sure you're back in the lvmdrp directory.
+3. Install the DRP package in the current python environment (see [contributing](#contributing-to-lvm-drp-development) section below for a replacement of this step). Make sure you're back in the lvmdrp directory.
 
     ```bash
     cd lvmdrp

cextern/README.txt

@@ -1,8 +1,7 @@
-C code to be compiled on setup goes here.
+C++ code to generate fast_median extension goes here.
 
-fast_median.cpp is a very fast median filter for 1d and 2d ndarrays
-with float or double data. The glue code is in fast_median.py
-
-compile using the Makefile in src/
+fast_median.cpp, fast_median.hpp: a very fast median filter for 1d and 2d ndarrays
+with float or double data.
+python/lvmdrp/external/fast_median.py: python interface for fast_median extension
 
 

tests/external/test_fast_median.py

@@ -0,0 +1,40 @@
+import numpy as np
+from lvmdrp.external.fast_median import fast_median_filter_2d, fast_median_filter_1d
+from scipy.ndimage import median_filter
+import time
+import pytest
+
+
+def test_median_filter_1d_speed():
+    np.random.seed(123)
+    img = np.random.random(size=100)*3 + 1000
+    img = img.astype(np.float32)
+
+    t = time.time()
+    img_fm = fast_median_filter_1d(img, size=5)
+    dt_fm = time.time()-t
+
+    t = time.time()
+    img_om = median_filter(img, size=5)
+    dt_om = time.time()-t
+
+    assert dt_om > dt_fm
+    assert img_om == pytest.approx(img_fm, rel=0.01)
+
+
+def test_median_filter_2d_speed():
+    np.random.seed(456)
+    img = np.random.random(size=[100,100])*3 + 1000
+    img = img.astype(np.float32)
+
+    t = time.time()
+    img_fm = fast_median_filter_2d(img, size=(5,5))
+    dt_fm = time.time()-t
+
+    t = time.time()
+    img_om = median_filter(img, size=(5,5))
+    dt_om = time.time()-t
+
+    assert dt_om > dt_fm
+    assert img_om == pytest.approx(img_fm, rel=0.01)
+