Merge pull request #18 from angus-g/testing

Add testing framework

.travis.yml

@@ -2,6 +2,12 @@ language: python
 cache: pip
 install:
   - pip install black
-  - pip install -e '.'
-script:
-  - black --check --verbose .
+  - pip install -r requirements.txt
+  - pip install -e '.[build]'
+jobs:
+  include:
+    - stage: "Code Formatting"
+      script: black --check --verbose .
+    - stage: "Tests"
+      name: "Unit Tests"
+      script: pytest

setup.py

@@ -7,4 +7,5 @@
     use_scm_version=True,
     setup_requires=["setuptools_scm"],
     packages=find_packages(),
+    extras_require={"build": ["pytest"]},
 )

test/test_filtering.py

@@ -0,0 +1,97 @@
+import pytest
+
+import numpy as np
+from scipy import signal
+import xarray as xr
+
+import filtering
+
+
+def velocity_series(nt, U0, f):
+    """Construct a 1D velocity timeseries."""
+
+    t = np.arange(nt) + 1
+    t0 = nt // 2 + 1  # middle time index
+    u = U0 + (U0 / 2) * np.sin(2 * np.pi * f * (t - t0))
+
+    return t, u
+
+
+def test_sanity():
+    """Sanity check of filtering.
+
+    Set up a mean velocity field with an oscillating component,
+    then filter out the mean.
+    """
+
+    # construct sample times (hrs) and velocity field (m/hr)
+    U0 = 100 / 24
+    w = 1 / 6  # tidal frequency
+    nt = 37
+    _, u = velocity_series(nt, U0, w)
+    assert u[nt // 2] == pytest.approx(U0)
+
+    # construct filter
+    f = signal.butter(4, w / 2, "highpass")
+    fu = signal.filtfilt(*f, u)
+    assert fu[nt // 2] == pytest.approx(0.0, abs=1e-2)
+
+
+def test_sanity_filtering_from_dataset(tmp_path):
+    """Sanity check of filtering using the library.
+
+    As with the :func:`~test_sanity` test, this sets up a mean
+    velocity field (in 2D) with an oscillating component. Because the
+    velocity field is uniform in time, the Lagrangian timeseries
+    should be the same as the 1D timeseries.
+    """
+
+    U0 = 100 / 24
+    w = 1 / 6
+    nt = 37
+    t, u = velocity_series(nt, U0, w)
+
+    # convert hours to seconds
+    u /= 3600
+    t *= 3600
+
+    x = np.array([0, 500, 1000])
+    y = np.array([0, 500, 1000])
+
+    # broadcast velocity to right shape
+    u_full = np.empty((nt, y.size, x.size))
+    u_full[:] = u[:, None, None]
+
+    # create dataset
+    d = xr.Dataset(
+        {
+            "u": (["time", "y", "x"], u_full),
+            "v": (["time", "y", "x"], np.zeros_like(u_full)),
+        },
+        coords={"x": x, "y": y, "time": t},
+    )
+
+    # dump to file
+    p = tmp_path / "data.nc"
+    d.to_netcdf(p)
+
+    f = filtering.LagrangeFilter(
+        "sanity_test",
+        {"U": str(p), "V": str(p)},
+        {"U": "u", "V": "v"},
+        {"lon": "x", "lat": "y", "time": "time"},
+        sample_variables=["U"],
+        mesh="flat",
+        window_size=17 * 3600,
+        highpass_frequency=(w / 2) / 3600,
+        advection_dt=30 * 60,
+    )
+
+    # filter from the middle of the series
+    filtered = f.filter_step(t[nt // 2])["var_U"]
+    # we expect a lot of parcels to hit the edge and die
+    # but some should stay alive
+    filtered = filtered[~np.isnan(filtered)]
+    assert filtered.size > 0
+    value = filtered.item(0)
+    assert value == pytest.approx(0.0, abs=1e-3)