interpolation support at commandline and predict_model levels

src/LarvaDatasets.jl

@@ -250,6 +250,7 @@ end
 function write_larva_dataset_hdf5(path, counts, files, refs, nsteps_before, nsteps_after;
         fixmwt=false, frameinterval=nothing,
     )
+    fixmwt && @warn "`fixmwt=true` is no longer supported"
     # this method mutates argument `refs`
     refs′= Tuple{Int, Int, Int, eltype(keys(counts))}[]
     for (label, count) in pairs(counts)
@@ -278,6 +279,9 @@ function write_larva_dataset_hdf5(path, counts, files, refs, nsteps_before, nste
         # extension
         h5["labels"] = collect(keys(counts))
         #h5["files"] = [f.source for f in files]
+        if !isnothing(frameinterval)
+            attributes(h5["samples"])["frame_interval"] = frameinterval
+        end
     end
 end
 

src/taggingbackends/explorer.py

@@ -415,7 +415,8 @@ run `poetry add {pkg}` from directory: \n
         return input_files, labels
 
     def generate_dataset(self, input_files,
-            labels=None, window_length=20, sample_size=None):
+            labels=None, window_length=20, sample_size=None,
+            frame_interval=None):
         """
         Generate a *larva_dataset hdf5* file in data/interim/{instance}/
         """
@@ -424,7 +425,8 @@ run `poetry add {pkg}` from directory: \n
                 input_files if isinstance(input_files, list) else str(input_files),
                 window_length,
                 labels=labels,
-                sample_size=sample_size)
+                sample_size=sample_size,
+                frameinterval=frame_interval)
 
     def compile_trxmat_database(self, input_dir,
             labels=None, window_length=20, sample_size=None, reuse_h5files=False):

src/taggingbackends/features/skeleton.py

@@ -19,3 +19,46 @@ def get_5point_spines(spine):
         return spine
     else:
         raise NotImplementedError(spine.shape)
+
+def interpolate(times, spines, anchor, window_length,
+                spine_interpolation='linear', frame_interval=0.1, **kwargs):
+    """
+    Interpolate spine series around anchor time `times[anchor]`, with about
+    `window_length // 2` time steps before and after, evenly spaced by
+    `frame_interval`.
+
+    Only linear interpolation is supported for now.
+    """
+    # m = anchor
+    # n = m + window_length
+    # if n <= spines.shape[0]:
+    #     return spines[m:n,:]
+    # else:
+    #     return
+    assert spine_interpolation == 'linear'
+    tstart, anchor, tstop = times[0], times[anchor], times[-1]
+    istart = np.trunc((tstart - anchor) / frame_interval).astype(int)
+    istop = np.trunc((tstop - anchor) / frame_interval).astype(int)
+    nframes_before = window_length // 2
+    nframes_after = window_length - 1 - nframes_before
+    istart = max(-nframes_before, istart)
+    istop = min(nframes_after, istop)
+    if istop - istart + 1 < window_length:
+        return
+    grid = range(istart, istop+1)
+    series = []
+    for i in grid:
+        t = round((anchor + i * frame_interval) * 1e4) * 1e-4
+        inext = np.flatnonzero(t <= times)[0]
+        tnext, xnext = times[inext], spines[inext]
+        if tnext == t:
+            x = xnext
+        else:
+            assert 0 < inext
+            tprev, xprev = times[inext-1], spines[inext-1]
+            x = interp(xprev, xnext, (t - tprev) / (tnext - tprev))
+        series.append(x)
+    return np.stack(series, axis=0)
+
+def interp(x0, x1, alpha):
+    return (1 - alpha) * x0 + alpha * x1

src/taggingbackends/main.py

@@ -8,7 +8,7 @@ def help(_print=False):
 Usage:  tagging-backend [train|predict] --model-instance <name>
         tagging-backend train ... --labels <comma-separated-list>
         tagging-backend train ... --sample-size <N> --window-length <T>
-        tagging-backend train ... --trxmat-only --reuse-h5files
+        tagging-backend train ... --frame-interval <I>
         tagging-backend train ... --pretrained-model-instance <name>
         tagging-backend predict ... --skip-make-dataset
 
@@ -21,6 +21,10 @@ spines. If option `--sample-size` is passed, <N> time segments are sampled from
 the raw database. The total length 3*<T> of time segments is 60 per default
 (20 *past* points, 20 *present* points and 20 *future* points).
 
+If frame interval <I> is specified (in seconds), spine series are resampled and
+interpolated around each time segment anchor (center).
+
+**Deprecated**:
 Option `--trxmat-only` is suitable for large databases made of trx.mat files
 only. Intermediate HDF5 files are generated prior to counting the various
 behavior labels and sampling time segments in the database. These intermediate
@@ -48,7 +52,7 @@ def main(fun=None):
         train_or_predict = sys.argv[1]
         project_dir = model_instance = None
         input_files, labels = [], []
-        sample_size = window_length = None
+        sample_size = window_length = frame_interval = None
         trxmat_only = reuse_h5files = False
         skip_make_dataset = skip_build_features = False
         pretrained_model_instance = None
@@ -72,6 +76,9 @@ def main(fun=None):
             elif sys.argv[k] == "--window-length":
                 k = k + 1
                 window_length = sys.argv[k]
+            elif sys.argv[k] == "--frame-interval":
+                k = k + 1
+                frame_interval = sys.argv[k]
             elif sys.argv[k] == "--trxmat-only":
                 trxmat_only = True
             elif sys.argv[k] == "--reuse-h5files":
@@ -98,6 +105,8 @@ def main(fun=None):
                 make_dataset_kwargs["sample_size"] = sample_size
             if window_length:
                 make_dataset_kwargs["window_length"] = window_length
+            if frame_interval:
+                make_dataset_kwargs["frame_interval"] = frame_interval
             if trxmat_only:
                 make_dataset_kwargs["trxmat_only"] = True
                 if reuse_h5files:
@@ -125,6 +134,9 @@ def main(fun=None):
             elif key in ("sample_size", "window_length"):
                 if isinstance(val, str):
                     val = int(val)
+            elif key in ("frame_interval",):
+                if isinstance(val, str):
+                    val = float(val)
             elif key == "labels":
                 if isinstance(val, str):
                     val = val.split(',')