manage time in network

Author: AntSimi

examples/16_network/pet_follow_particle.py

@@ -133,12 +133,12 @@ def update(frame):
 shape = (n.obs.size, 2)
 # Forward run
 i_target_f, pct_target_f = -ones(shape, dtype="i4"), zeros(shape, dtype="i1")
-for t in range(t_start, t_end - dt):
+for t in arange(t_start, t_end - dt):
     particle_candidate(c, n, step, t, i_target_f, pct_target_f, n_days=dt)
 
 # Backward run
 i_target_b, pct_target_b = -ones(shape, dtype="i4"), zeros(shape, dtype="i1")
-for t in range(t_start + dt, t_end):
+for t in arange(t_start + dt, t_end):
     particle_candidate(c, n, step, t, i_target_b, pct_target_b, n_days=-dt)
 
 # %%

notebooks/python_module/16_network/pet_follow_particle.ipynb

@@ -120,7 +120,7 @@
       },
       "outputs": [],
       "source": [
-        "step = 1 / 60.0\n\nx, y = meshgrid(arange(24, 36, step), arange(31, 36, step))\nx0, y0 = x.reshape(-1), y.reshape(-1)\n# Pre-order to speed up\n_, i = group_obs(x0, y0, 1, 360)\nx0, y0 = x0[i], y0[i]\n\nt_start, t_end = n.period\ndt = 14\n\nshape = (n.obs.size, 2)\n# Forward run\ni_target_f, pct_target_f = -ones(shape, dtype=\"i4\"), zeros(shape, dtype=\"i1\")\nfor t in range(t_start, t_end - dt):\n    particle_candidate(x0, y0, c, n, t, i_target_f, pct_target_f, n_days=dt)\n\n# Backward run\ni_target_b, pct_target_b = -ones(shape, dtype=\"i4\"), zeros(shape, dtype=\"i1\")\nfor t in range(t_start + dt, t_end):\n    particle_candidate(x0, y0, c, n, t, i_target_b, pct_target_b, n_days=-dt)"
+        "step = 1 / 60.0\n\nx, y = meshgrid(arange(24, 36, step), arange(31, 36, step))\nx0, y0 = x.reshape(-1), y.reshape(-1)\n# Pre-order to speed up\n_, i = group_obs(x0, y0, 1, 360)\nx0, y0 = x0[i], y0[i]\n\nt_start, t_end = n.period\ndt = 14\n\nshape = (n.obs.size, 2)\n# Forward run\ni_target_f, pct_target_f = -ones(shape, dtype=\"i4\"), zeros(shape, dtype=\"i1\")\nfor t in arange(t_start, t_end - dt):\n    particle_candidate(x0, y0, c, n, t, i_target_f, pct_target_f, n_days=dt)\n\n# Backward run\ni_target_b, pct_target_b = -ones(shape, dtype=\"i4\"), zeros(shape, dtype=\"i1\")\nfor t in arange(t_start + dt, t_end):\n    particle_candidate(x0, y0, c, n, t, i_target_b, pct_target_b, n_days=-dt)"
       ]
     },
     {

notebooks/python_module/16_network/pet_segmentation_anim.ipynb

@@ -15,7 +15,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "\n# Network segmentation process\n"
+        "\nNetwork segmentation process\n============================\n"
       ]
     },
     {
@@ -62,7 +62,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "## Load data\nLoad data where observations are put in same network but no segmentation\n\n"
+        "Load data\n---------\nLoad data where observations are put in same network but no segmentation\n\n"
       ]
     },
     {
@@ -80,7 +80,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "## Do segmentation\nSegmentation based on maximum overlap, temporal window for candidates = 5 days\n\n"
+        "Do segmentation\n---------------\nSegmentation based on maximum overlap, temporal window for candidates = 5 days\n\n"
       ]
     },
     {
@@ -98,7 +98,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "## Anim\n\n"
+        "Anim\n----\n\n"
       ]
     },
     {
@@ -109,14 +109,14 @@
       },
       "outputs": [],
       "source": [
-        "def update(i_frame):\n    tr = TRACKS[i_frame]\n    mappable_tracks.set_array(tr)\n    s = 40 * ones(tr.shape)\n    s[tr == 0] = 4\n    mappable_tracks.set_sizes(s)\n\n    indices_frames = INDICES[i_frame]\n    mappable_CONTOUR.set_data(\n        e.contour_lon_e[indices_frames],\n        e.contour_lat_e[indices_frames],\n    )\n    mappable_CONTOUR.set_color(cmap.colors[tr[indices_frames] % len(cmap.colors)])\n    return (mappable_tracks,)\n\n\nfig = plt.figure(figsize=(16, 9), dpi=60)\nax = fig.add_axes([0.04, 0.06, 0.94, 0.88], projection=GUI_AXES)\nax.set_title(f\"{len(e)} observations to segment\")\nax.set_xlim(19, 29), ax.set_ylim(31, 35.5), ax.grid()\nvmax = TRACKS[-1].max()\ncmap = ListedColormap([\"gray\", *e.COLORS[:-1]], name=\"from_list\", N=vmax)\nmappable_tracks = ax.scatter(\n    e.lon, e.lat, c=TRACKS[0], cmap=cmap, vmin=0, vmax=vmax, s=20\n)\nmappable_CONTOUR = ax.plot(\n    e.contour_lon_e[INDICES[0]], e.contour_lat_e[INDICES[0]], color=cmap.colors[0]\n)[0]\nani = VideoAnimation(fig, update, frames=range(1, len(TRACKS), 4), interval=125)"
+        "def update(i_frame):\n    tr = TRACKS[i_frame]\n    mappable_tracks.set_array(tr)\n    s = 40 * ones(tr.shape)\n    s[tr == 0] = 4\n    mappable_tracks.set_sizes(s)\n\n    indices_frames = INDICES[i_frame]\n    mappable_CONTOUR.set_data(\n        e.contour_lon_e[indices_frames], e.contour_lat_e[indices_frames],\n    )\n    mappable_CONTOUR.set_color(cmap.colors[tr[indices_frames] % len(cmap.colors)])\n    return (mappable_tracks,)\n\n\nfig = plt.figure(figsize=(16, 9), dpi=60)\nax = fig.add_axes([0.04, 0.06, 0.94, 0.88], projection=GUI_AXES)\nax.set_title(f\"{len(e)} observations to segment\")\nax.set_xlim(19, 29), ax.set_ylim(31, 35.5), ax.grid()\nvmax = TRACKS[-1].max()\ncmap = ListedColormap([\"gray\", *e.COLORS[:-1]], name=\"from_list\", N=vmax)\nmappable_tracks = ax.scatter(\n    e.lon, e.lat, c=TRACKS[0], cmap=cmap, vmin=0, vmax=vmax, s=20\n)\nmappable_CONTOUR = ax.plot(\n    e.contour_lon_e[INDICES[0]], e.contour_lat_e[INDICES[0]], color=cmap.colors[0]\n)[0]\nani = VideoAnimation(fig, update, frames=range(1, len(TRACKS), 4), interval=125)"
       ]
     },
     {
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "## Final Result\n\n"
+        "Final Result\n------------\n\n"
       ]
     },
     {
@@ -147,7 +147,7 @@
       "name": "python",
       "nbconvert_exporter": "python",
       "pygments_lexer": "ipython3",
-      "version": "3.7.9"
+      "version": "3.7.7"
     }
   },
   "nbformat": 4,

src/py_eddy_tracker/observations/tracking.py

@@ -16,6 +16,7 @@
     degrees,
     empty,
     histogram,
+    int_,
     median,
     nan,
     ones,
@@ -601,6 +602,12 @@ def plot(self, ax, ref=None, **kwargs):
 
     def split_network(self, intern=True, **kwargs):
         """Return each group (network) divided in segments"""
+        # Find timestep of dataset
+        # FIXME : how to know exact time sampling
+        t = unique(self.time)
+        dts = t[1:] - t[:-1]
+        timestep = median(dts)
+
         track_s, track_e, track_ref = build_index(self.tracks)
         ids = empty(
             len(self),
@@ -614,7 +621,7 @@ def split_network(self, intern=True, **kwargs):
                 ("next_obs", "i4"),
             ],
         )
-        ids["group"], ids["time"] = self.tracks, self.time
+        ids["group"], ids["time"] = self.tracks, int_(self.time / timestep)
         # Initialisation
         # To store the id of the segments, the backward and forward cost associations
         ids["track"], ids["previous_cost"], ids["next_cost"] = 0, 0, 0
@@ -641,6 +648,7 @@ def split_network(self, intern=True, **kwargs):
             local_ids["next_obs"][m] += i_s
         if display_iteration:
             print()
+        ids["time"] *= timestep
         return ids
 
     def set_tracks(self, x, y, ids, window, **kwargs):
@@ -652,8 +660,7 @@ def set_tracks(self, x, y, ids, window, **kwargs):
         :param ndarray ids: several fields like time, group, ...
         :param int windows: number of days where observations could missed
         """
-
-        time_index = build_index(ids["time"])
+        time_index = build_index((ids["time"]).astype("i4"))
         nb = x.shape[0]
         used = zeros(nb, dtype="bool")
         track_id = 1
@@ -698,8 +705,7 @@ def get_previous_obs(
         i_current, ids, x, y, time_s, time_e, time_ref, window, **kwargs
     ):
         """Backward association of observations to the segments"""
-
-        time_cur = ids["time"][i_current]
+        time_cur = int_(ids["time"][i_current])
         t0, t1 = time_cur - 1 - time_ref, max(time_cur - window - time_ref, 0)
         for t_step in range(t0, t1 - 1, -1):
             i0, i1 = time_s[t_step], time_e[t_step]
@@ -729,7 +735,7 @@ def get_previous_obs(
     def get_next_obs(i_current, ids, x, y, time_s, time_e, time_ref, window, **kwargs):
         """Forward association of observations to the segments"""
         time_max = time_e.shape[0] - 1
-        time_cur = ids["time"][i_current]
+        time_cur = int_(ids["time"][i_current])
         t0, t1 = time_cur + 1 - time_ref, min(time_cur + window - time_ref, time_max)
         if t0 > time_max:
             return -1