-
Notifications
You must be signed in to change notification settings - Fork 55
Home
AntSimi edited this page Jul 3, 2020
·
5 revisions
You can create your own tracking procedure with inheritance of py_eddy_tracker.observations.EddiesObservations
To do this you simply create a new class in a python file
from py_eddy_tracker.observations.observation import EddiesObservations as Model
class MyTracking(Model):
def mask_function(self, other, distance):
"""The role of this function, it's to reduce the number of case to investigate
self : observation from timestamp t
other : observation from the next timestamp
distance : matrix of distance between self observation and other observation
"""
# We keep only association under 150 km
mask = distance < 150
return mask
@classmethod
def cost_function(cls, *args, **kwargs):
"""This method will be call after mask_function
args will contains :
observation from t which must be compare with observation from next timestep
"""
# Compute intersection between two polygons and return ratio of intersection on area of first polygon
values = cls.cost_function_common_area(*args, **kwargs)
# 1 - poly_in.intersection(poly_out).area / poly_in.area
# we mask tiny intersection
m = values > .9
values.mask += m
return values
In order to use this new tracking method the new file must be in your PYTHONPATH, and you need to specify the following in your yaml file:
PATHS:
# Files produces with EddyIdentification
FILES_PATTERN: {files_in}
SAVE_DIR: {output_path}
CLASS:
MODULE: my_filename
CLASS: MyTracking
Run this in the command line as: $ EddyTracking my_tracking.yaml
Common functions, which can be overwritten:
class MyTracking(Model):
@staticmethod
def propagate(previous_obs, current_obs, obs_to_extend, dead_track, nb_next, model):
"""
In order to filled virtual obs (C)
Args:
previous_obs: previous obs from current (A)
current_obs: previous obs from virtual (B)
Returns:
New position C = B + AB
"""
def mask_function(self, other, distance):
"""In order to do a preselection on candidate to match, it must be lowcost
Some basic geographic function are available:
- shifted_ellipsoid_degrees_mask
- fixed_ellipsoid_mask
"""
def solve_function(self, cost_matrix):
"""Select match with the cost_matrix, in order to solve conflict
return a list of match
two function are already usable : solve_simultaneous, solve_first
"""
def post_process_link(self, other, i_self, i_other):
"""Do some selection on the list of match
When two self obs use the same other obs ? what did you do?
"""