v0.10.0.md

@@ -20,6 +20,33 @@
 - The differences are just due to the difference in floating point storage between
   SQL and JSON. I believe this difference is small enough to be negligible
 
+# npy file format
+
+## Related MRs
+
+* https://git.ligo.org/lscsoft/pesummary/-/merge_requests/474
+
+## Test
+
+To check that posterior samples can both be written to `npy` and read back in using both the pesummary's `io` module and the standard numpy `load` function, we ran the following test:
+
+```python
+from pesummary.io import read
+import numpy as np
+
+f = read(PATH)
+f.write(file_format="numpy", filename="example_output.npy", outdir="./")
+g = read("./example_output.npy")
+assert sorted(f.parameters) == sorted(g.parameters)
+np.testing.assert_almost_equal(np.array(f.samples), np.array(g.samples))
+
+numpy = np.load("./example_output.npy")
+assert sorted(f.parameters) == sorted(list(numpy.dtype.names))
+np.testing.assert_almost_equal(np.array(f.samples), np.array([list(i) for i in numpy]))
+```
+
+We ran this with `PATH` indicating the path to a `LALInference hdf5` file and a `bilby json` file and worked as expected.
+
 # CSV file format
 
 ## Related MRs