Merge Split algorithm added

tobac/merge_split.py

@@ -0,0 +1,345 @@
+#Tobac merge and split v0.1
+
+from geopy.distance import geodesic
+from networkx import *
+import numpy as np
+from pandas.core.common import flatten
+import xarray as xr
+
+def compress_all(nc_grids, min_dims=2):
+    for var in nc_grids:
+        if len(nc_grids[var].dims) >= min_dims:
+            # print("Compressing ", var)
+            nc_grids[var].encoding["zlib"] = True
+            nc_grids[var].encoding["complevel"] = 4
+            nc_grids[var].encoding["contiguous"] = False
+    return nc_grids
+
+
+def standardize_track_dataset(TrackedFeatures, Mask, Projection):
+    """ Combine a feature mask with the feature data table into a common dataset.
+    Also renames th
+
+    returned by tobac.themes.tobac_v1.segmentation
+    with the TrackedFeatures dataset returned by tobac.themes.tobac_v1.linking_trackpy.
+
+    Also rename the variables to be more desciptive and comply with cf-tree.
+
+    Convert the default cell parent ID  to an integer table.
+
+    Add a cell dimension to reflect
+
+    Projection is an xarray DataArray
+
+    TODO: Add metadata attributes to
+
+     """
+    feature_standard_names = {
+        # new variable name, and long description for the NetCDF attribute
+        'frame':('feature_time_index',
+            'positional index of the feature along the time dimension of the mask, from 0 to N-1'),
+        'hdim_1':('feature_hdim1_coordinate',
+            'position of the feature along the first horizontal dimension in grid point space; a north-south coordinate for dim order (time, y, x).'
+        'The numbering is consistent with positional indexing of the coordinate, but can be'
+            'fractional, to account for a centroid not aligned to the grid.'),
+        'hdim_2':('feature_hdim2_coordinate',
+            'position of the feature along the second horizontal dimension in grid point space; an east-west coordinate for dim order (time, y, x)'
+            'The numbering is consistent with positional indexing of the coordinate, but can be'
+            'fractional, to account for a centroid not aligned to the grid.'),
+        'idx':('feature_id_this_frame',),
+        'num':('feature_grid_cell_count',
+            'Number of grid points that are within the threshold of this feature'),
+        'threshold_value':('feature_threshold_max',
+            "Feature number within that frame; starts at 1, increments by 1 to the number of features for each frame, and resets to 1 when the frame increments"),
+        'feature':('feature_id',
+            "Unique number of the feature; starts from 1 and increments by 1 to the number of features"),
+        'time':('feature_time','time of the feature, consistent with feature_time_index'),
+        'timestr':('feature_time_str','String representation of the feature time, YYYY-MM-DD HH:MM:SS'),
+        'projection_y_coordinate':('feature_projection_y_coordinate','y position of the feature in the projection given by ProjectionCoordinateSystem'),
+        'projection_x_coordinate':('feature_projection_x_coordinate','x position of the feature in the projection given by ProjectionCoordinateSystem'),
+        'lat':('feature_latitude','latitude of the feature'),
+        'lon':('feature_longitude','longitude of the feature'),
+        'ncells':('feature_ncells','number of grid cells for this feature (meaning uncertain)'),
+        'areas':('feature_area',),
+        'isolated':('feature_isolation_flag',),
+        'num_objects':('number_of_feature_neighbors',),
+        'cell':('feature_parent_cell_id',),
+        'time_cell':('feature_parent_cell_elapsed_time',),
+        'segmentation_mask':('2d segmentation mask',)
+    }
+    new_feature_var_names = {k:feature_standard_names[k][0] for k in feature_standard_names.keys()
+                             if k in TrackedFeatures.variables.keys()}
+
+    TrackedFeatures = TrackedFeatures.drop(['cell_parent_track_id'])
+    # Combine Track and Mask variables. Use the 'feature' variable as the coordinate variable instead of
+    # the 'index' variable and call the dimension 'feature'
+    ds = xr.merge([TrackedFeatures.swap_dims({'index':'feature'}).drop('index').rename_vars(new_feature_var_names),
+                  Mask])
+
+    # Add the projection data back in
+    ds['ProjectionCoordinateSystem']=Projection
+
+    # Convert the cell ID variable from float to integer
+    if 'int' not in str(TrackedFeatures.cell.dtype):
+        # The raw output from the tracking is actually an object array
+        # array([nan, 2, 3], dtype=object)
+        # (and is cast to a float array when saved as NetCDF, I think).
+        # Cast to float.
+        int_cell = xr.zeros_like(TrackedFeatures.cell, dtype='int64')
+
+        cell_id_data = TrackedFeatures.cell.astype('float64')
+        valid_cell = np.isfinite(cell_id_data)
+        valid_cell_ids = cell_id_data[valid_cell]
+        if not (np.unique(valid_cell_ids) > 0).all():
+            raise AssertionError('Lowest cell ID cell is less than one, conflicting with use of zero to indicate an untracked cell')
+        int_cell[valid_cell] = valid_cell_ids.astype('int64')
+        #ds['feature_parent_cell_id'] = int_cell
+    return ds
+
+
+
+def merge_split(TRACK,distance = 25000,frame_len = 5):
+    '''
+    function to  postprocess tobac track data for merge/split cells
+    Input:
+        TRACK:    xarray dataset of tobac Track information
+
+        distance:    float, optional distance threshold prior to adding a pair of points
+                            into the minimum spanning tree. Default is 25000 meters.
+
+        frame_len: float, optional threshold for the spanning length within which two points
+                          can be separated. Default is five (5) frames.
+
+
+    Output:
+        d:    xarray dataset of 
+                        feature position along 1st horizontal dimension
+        hdim2_index:    float
+                        feature position along 2nd horizontal dimension
+
+    Example:
+        d = merge_split(Track)
+       ds = standardize_track_dataset(Track, refl_mask, data['ProjectionCoordinateSystem'])
+        # both_ds = xarray.combine_by_coords((ds,d), compat='override')
+        both_ds = xr.merge([ds, d],compat ='override')
+        both_ds = compress_all(both_ds)
+        both_ds.to_netcdf(os.path.join(savedir,'Track_features_merges.nc'))
+
+    '''
+    track_groups = TRACK.groupby('cell')
+    cell_ids = {cid:len(v) for cid, v in track_groups.groups.items()}
+    id_data = np.fromiter(cell_ids.keys(), dtype=int)
+    count_data = np.fromiter(cell_ids.values(), dtype=int)
+    all_frames = np.sort(np.unique(TRACK.frame))
+    a_points = list()
+    b_points = list()
+    a_names = list()
+    b_names = list()
+    dist = list()
+
+
+    for i in id_data:
+        #print(i)
+        a_pointx = track_groups[i].projection_x_coordinate[-1].values
+        a_pointy = track_groups[i].projection_y_coordinate[-1].values
+        for j in id_data:
+            b_pointx = track_groups[j].projection_x_coordinate[0].values
+            b_pointy = track_groups[j].projection_y_coordinate[0].values
+            d = np.linalg.norm(np.array((a_pointx, a_pointy)) - np.array((b_pointx, b_pointy)))
+            if d <= distance:
+                a_points.append([a_pointx,a_pointy])
+                b_points.append([b_pointx, b_pointy])
+                dist.append(d)
+                a_names.append(i)
+                b_names.append(j)
+
+
+
+
+
+#     for i in id_data:
+#         a_pointx = track_groups[i].grid_longitude[-1].values
+#         a_pointy = track_groups[i].grid_latitude[-1].values
+#         for j in id_data:
+#             b_pointx = track_groups[j].grid_longitude[0].values
+#             b_pointy = track_groups[j].grid_latitude[0].values
+#             d = geodesic((a_pointy,a_pointx),(b_pointy,b_pointx)).km
+#             if d <= distance:
+#                 a_points.append([a_pointx,a_pointy])
+#                 b_points.append([b_pointx, b_pointy])
+#                 dist.append(d)
+#                 a_names.append(i)
+#                 b_names.append(j)
+
+    id = []
+    for i in range(len(dist)-1, -1, -1): 
+        if a_names[i] == b_names[i]:
+            id.append(i)
+            a_points.pop(i)
+            b_points.pop(i)
+            dist.pop(i)
+            a_names.pop(i)
+            b_names.pop(i)
+        else:
+            continue
+
+    g = Graph()
+    for i in np.arange(len(dist)):
+        g.add_edge(a_names[i], b_names[i],weight=dist[i])
+
+    tree = minimum_spanning_edges(g)
+    tree_list = list(minimum_spanning_edges(g))
+
+    new_tree = []
+    for i,j in enumerate(tree_list):
+        frame_a = np.nanmax(track_groups[j[0]].frame.values)
+        frame_b = np.nanmin(track_groups[j[1]].frame.values)
+        if np.abs(frame_a - frame_b) <= frame_len:
+            new_tree.append(tree_list[i][0:2])
+    new_tree_arr = np.array(new_tree)
+
+    TRACK['cell_parent_track_id'] = np.zeros(len(TRACK['cell'].values))
+    cell_id = np.unique(TRACK.cell.values.astype(float)[~np.isnan(TRACK.cell.values.astype(float))])
+    track_id = dict() #same size as number of total merged tracks
+
+    arr = np.array([0])
+    for p in cell_id:
+        j = np.where(arr == int(p))
+        if len(j[0]) > 0:
+            continue
+        else:
+            k = np.where(new_tree_arr == p)
+            if len(k[0]) == 0:
+                track_id[p] = [p]
+                arr = np.append(arr,p)
+            else:
+                temp1 = list(np.unique(new_tree_arr[k[0]]))
+                temp = list(np.unique(new_tree_arr[k[0]]))
+
+                for l in range(15):
+                    for i in temp1:
+                        k2 = np.where(new_tree_arr == i)
+                        temp.append(list(np.unique(new_tree_arr[k2[0]]).squeeze()))
+                        temp = list(flatten(temp))
+                        temp = list(np.unique(temp))
+
+                    if len(temp1) == len(temp):
+                        break
+                    temp1 = np.array(temp)
+
+                for i in temp1:
+                    k2 = np.where(new_tree_arr == i)
+                    temp.append(list(np.unique(new_tree_arr[k2[0]]).squeeze()))
+
+                temp = list(flatten(temp))
+                temp = list(np.unique(temp))
+                arr = np.append(arr,np.unique(temp))
+
+                track_id[np.nanmax(np.unique(temp))] = list(np.unique(temp))
+
+
+
+    storm_id = [0] #default because we don't track larger storm systems *yet*
+    print('found storm id')
+
+
+    track_parent_storm_id = np.repeat(0, len(track_id)) #This will always be zero when we don't track larger storm systems *yet*
+    print('found track parent storm ids')
+
+    track_ids = np.array(list(track_id.keys()))
+    print('found track ids')
+
+
+    cell_id = list(np.unique(TRACK.cell.values.astype(float)[~np.isnan(TRACK.cell.values.astype(float))]))
+    print('found cell ids')
+
+    cell_parent_track_id = []
+
+    for i, id in enumerate(track_id):
+
+        if len(track_id[int(id)]) ==  1:
+            cell_parent_track_id.append(int(id))
+
+        else:
+            cell_parent_track_id.append(np.repeat(int(id),len(track_id[int(id)])))
+
+
+    cell_parent_track_id = list(flatten(cell_parent_track_id))
+    print('found cell parent track ids')
+
+    feature_parent_cell_id = list(TRACK.cell.values.astype(float))
+
+    print('found feature parent cell ids')
+
+    #This version includes all the feature regardless of if they are used in cells or not.
+    feature_id = list(TRACK.feature.values.astype(int))
+    print('found feature ids')
+
+    feature_parent_storm_id = np.repeat(0,len(feature_id)) #we don't do storms atm
+    print('found feature parent storm ids')
+
+    feature_parent_track_id  = []
+    feature_parent_track_id = np.zeros(len(feature_id))
+    for i, id in enumerate(feature_id):
+        cellid = feature_parent_cell_id[i]
+        if np.isnan(cellid):
+            feature_parent_track_id[i] = -1 
+        else:
+            j = np.where(cell_id == cellid)
+            j = np.squeeze(j)
+            trackid = cell_parent_track_id[j]
+            feature_parent_track_id[i] = trackid
+
+    print('found feature parent track ids')
+
+
+    storm_child_track_count = [len(track_id)]
+    print('found storm child track count')
+
+    track_child_cell_count = []
+    for i,id in enumerate(track_id):
+        track_child_cell_count.append(len(track_id[int(id)]))
+    print('found track child cell count')
+
+
+    cell_child_feature_count = []
+    for i,id in enumerate(cell_id):
+        cell_child_feature_count.append(len(track_groups[id].feature.values))
+    print('found cell child feature count')
+
+    storm_child_cell_count =  [len(cell_id)] 
+    storm_child_feature_count = [len(feature_id)]
+
+    storm_dim = 'nstorms'
+    track_dim = 'ntracks'
+    cell_dim = 'ncells'
+    feature_dim = 'nfeatures'
+
+    d = xr.Dataset({
+        'storm_id': (storm_dim, storm_id),
+        'track_id': (track_dim, track_ids),
+        'track_parent_storm_id': (track_dim, track_parent_storm_id),
+        'cell_id': (cell_dim, cell_id),
+        'cell_parent_track_id': (cell_dim, cell_parent_track_id),
+        'feature_id': (feature_dim, feature_id),
+        'feature_parent_cell_id': (feature_dim, feature_parent_cell_id),
+        'feature_parent_track_id': (feature_dim, feature_parent_track_id),
+        'feature_parent_storm_id': (feature_dim, feature_parent_storm_id),
+        'storm_child_track_count': (storm_dim, storm_child_track_count),
+        'storm_child_cell_count': (storm_dim, storm_child_cell_count),
+        'storm_child_feature_count': (storm_dim, storm_child_feature_count),
+        'track_child_cell_count': (track_dim, track_child_cell_count),
+        'cell_child_feature_count': (cell_dim, cell_child_feature_count),
+            })
+    d = d.set_coords(['feature_id','cell_id', 'track_id', 'storm_id']) 
+
+    assert len(track_id) == len(track_parent_storm_id)
+    assert len(cell_id) == len(cell_parent_track_id)
+    assert len(feature_id) == len(feature_parent_cell_id)
+    assert sum(storm_child_track_count) == len(track_id)
+    assert sum(storm_child_cell_count) == len(cell_id)
+    assert sum(storm_child_feature_count) == len(feature_id)
+    assert sum(track_child_cell_count) == len(cell_id)
+    assert sum([sum(cell_child_feature_count),(len(np.where(feature_parent_track_id < 0)[0]))]) == len(feature_id)
+
+    return d