else:
ids["previous_obs"][i_next_] = i_next
i_next = i_next_
def get_obs(dataset):
"Function to isolate a specific obs"
return np.where((dataset.lat > 33) * (dataset.lat < 34) *
(dataset.lon > 22) * (dataset.lon < 23) *
(dataset.time > 20630) * (dataset.time < 20650))[0][0]
# %%
# Get original network, we will isolate only relative at order *2*
n = NetworkObservations.load_file(
"/data/adelepoulle/work/Eddies/20201217_network_build/tracking/med/Anticyclonic_seg.nc"
)
n = n.extract_with_mask(n.track == n.track[get_obs(n)])
n_ = n.relative(get_obs(n), order=2)
# %%
# Display the default segmentation
ax = start_axes(n_.infos())
n_.plot(ax, color_cycle=n.COLORS)
update_axes(ax)
fig = plt.figure(figsize=(15, 5))
ax = fig.add_axes([0.04, 0.05, 0.92, 0.92])
ax.xaxis.set_major_formatter(formatter), ax.grid()
_ = n_.display_timeline(ax)
@staticmethod
def get_next_obs(i_current, ids, x, y, time_s, time_e, time_ref, window, **kwargs):
TRACKS.append(ids["track"].copy())
INDICES.append(i_current)
return TrackEddiesObservations.get_next_obs(
i_current, ids, x, y, time_s, time_e, time_ref, window, **kwargs
)
# %%
# Load data
# ---------
# Load data where observations are put in same network but no segmentation
# Get a known network for the demonstration
n = NetworkObservations.load_file(get_demo_path("network_med.nc")).network(651)
# We keep only some segment
n = n.relative(get_obs(n), order=2)
print(len(n))
# We convert and order object like segmentation was never happen on observations
e = n.astype(MyTrack)
e.obs.sort(order=("track", "time"), kind="stable")
# %%
# Do segmentation
# ---------------
# Segmentation based on maximum overlap, temporal window for candidates = 5 days
matrix = e.split_network(intern=False, window=5)
# %%
i_next_ = -1
else:
ids["previous_obs"][i_next_] = i_next
i_next = i_next_
def get_obs(dataset):
"Function to isolate a specific obs"
return where((dataset.lat > 33) * (dataset.lat < 34) * (dataset.lon > 22) *
(dataset.lon < 23) * (dataset.time > 20630) *
(dataset.time < 20650))[0][0]
# %%
# Get original network, we will isolate only relative at order *2*
n = NetworkObservations.load_file(get_demo_path("network_med.nc")).network(651)
n_ = n.relative(get_obs(n), order=2)
# %%
# Display the default segmentation
ax = start_axes(n_.infos())
n_.plot(ax, color_cycle=n.COLORS)
update_axes(ax)
fig = plt.figure(figsize=(15, 5))
ax = fig.add_axes([0.04, 0.05, 0.92, 0.92])
ax.xaxis.set_major_formatter(formatter), ax.grid()
_ = n_.display_timeline(ax)
# %%
# Run a new segmentation
# ----------------------
# In this example some fields are removed (effective_contour_longitude,...) in order to save time for doc building
eddies_tracks.field_table()
# %%
# Networks
# --------
# Network files use some specific fields :
#
# - track : ID of network (ID 0 correspond to lonely eddies)
# - segment : ID of a segment within a network (from 1 to N)
# - previous_obs : Index of the previous observation in the full dataset,
# if -1 there are no previous observation (the segment starts)
# - next_obs : Index of the next observation in the full dataset, if -1 there are no next observation (the segment ends)
# - previous_cost : Result of the cost function (1 is a good association, 0 is bad) with previous observation
# - next_cost : Result of the cost function (1 is a good association, 0 is bad) with next observation
eddies_network = NetworkObservations.load_file(get_demo_path("network_med.nc"))
eddies_network.field_table()
# %%
sl = slice(70, 100)
Table(
eddies_network.network(651).obs[sl][[
"time",
"track",
"segment",
"previous_obs",
"previous_cost",
"next_obs",
"next_cost",
]])
"""
Network Analysis
================
"""
from matplotlib import pyplot as plt
from numpy import ma
import py_eddy_tracker.gui
from py_eddy_tracker.data import get_remote_sample
from py_eddy_tracker.observations.network import NetworkObservations
n = NetworkObservations.load_file(
get_remote_sample(
"eddies_med_adt_allsat_dt2018_err70_filt500_order1/Anticyclonic_network.nc"
))
# %%
# Parameters
step = 1 / 10.0
bins = ((-10, 37, step), (30, 46, step))
kw_time = dict(cmap="terrain_r", factor=100.0 / n.nb_days, name="count")
kw_ratio = dict(cmap=plt.get_cmap("YlGnBu_r", 10))
# %%
# Functions
def start_axes(title):
fig = plt.figure(figsize=(13, 5))
ax = fig.add_axes([0.03, 0.03, 0.90, 0.94], projection="full_axes")
ax.set_xlim(-6, 36.5), ax.set_ylim(30, 46)
ax.set_aspect("equal")
ax.set_title(title, weight="bold")
from py_eddy_tracker.observations.tracking import TrackEddiesObservations
# %%
# Load data
# ---------
# Load data where observations are put in same network but no segmentation
e = TrackEddiesObservations.load_file(data.get_path("c568803.nc"))
# FIXME : Must be rewrote
e.lon[:] = (e.lon + 180) % 360 - 180
e.contour_lon_e[:] = ((e.contour_lon_e.T - e.lon + 180) % 360 - 180 + e.lon).T
e.contour_lon_s[:] = ((e.contour_lon_s.T - e.lon + 180) % 360 - 180 + e.lon).T
# %%
# Do segmentation
# ---------------
# Segmentation based on maximum overlap, temporal window for candidates = 5 days
n = NetworkObservations.from_split_network(e, e.split_network(intern=False, window=5))
# %%
# Timeline
# --------
# %%
# Display timeline with events
# A segment generated by a splitting is marked with a star
#
# A segment merging in another is marked with an exagon
fig = plt.figure(figsize=(15, 5))
ax = fig.add_axes([0.04, 0.04, 0.92, 0.92])
_ = n.display_timeline(ax)
# %%
fig = plt.figure(figsize=(15, 5))
ax = fig.add_axes([0.03, 0.06, 0.90, 0.88])
ax.set_title(title, weight="bold")
ax.xaxis.set_major_formatter(formatter), ax.grid()
return ax
def update_axes(ax, mappable=None):
ax.grid(True)
if mappable:
return plt.colorbar(mappable, cax=ax.figure.add_axes([0.94, 0.05, 0.01, 0.9]))
# %%
# We know the network ID, we will get directly
ioannou_case = NetworkObservations.load_file(get_path("network_med.nc")).network(651)
print(ioannou_case.infos())
# %%
# It seems that this network is huge! Our case is visible at 22E 33.5N
ax = start_axes()
ioannou_case.plot(ax, color_cycle=ioannou_case.COLORS)
update_axes(ax)
# %%
# Full Timeline
# -------------
# The network span for many years... How to cut the interesting part?
fig = plt.figure(figsize=(15, 5))
ax = fig.add_axes([0.04, 0.05, 0.92, 0.92])
ax.xaxis.set_major_formatter(formatter), ax.grid()
ax.xaxis.set_major_formatter(formatter), ax.grid()
return ax
def update_axes(ax, mappable=None):
ax.grid(True)
if mappable:
return plt.colorbar(mappable,
cax=ax.figure.add_axes([0.94, 0.05, 0.01, 0.9]))
# %%
# We know the position and the time of a specific eddy
#
# `n.extract_with_mask` give us the corresponding network
n = NetworkObservations.load_file("med/Anticyclonic_seg.nc")
i = np.where((n.lat > 33) * (n.lat < 34) * (n.lon > 22) * (n.lon < 23) *
(n.time > 20630) * (n.time < 20650))[0][0]
ioannou_case = n.extract_with_mask(n.track == n.track[i])
print(ioannou_case.infos())
# %%
# It seems that this network is huge! Our case is in purple...
ax = start_axes()
ioannou_case.plot(ax)
update_axes(ax)
# %%
# Full Timeline
# -------------
# The network span for many years... How to cut the interesting part?
"""
Network basic manipulation
==========================
"""
from matplotlib import pyplot as plt
from numpy import where
import py_eddy_tracker.gui
from py_eddy_tracker import data
from py_eddy_tracker.observations.network import NetworkObservations
# %%
# Load data
# ---------
# Load data where observations are put in same network but no segmentation
n = NetworkObservations.load_file(data.get_path("network_med.nc")).network(651)
i = where((n.lat > 33) * (n.lat < 34) * (n.lon > 22) * (n.lon < 23) *
(n.time > 20630) * (n.time < 20650))[0][0]
# For event use
n2 = n.relative(i, order=2)
n = n.relative(i, order=4)
n.numbering_segment()
# %%
# Timeline
# --------
# %%
# Display timeline with events
# A segment generated by a splitting is marked with a star
#
pcolormesh = ax.pcolorfast(x_g_, y_g_, lavd, **kw_vorticity)
update_axes(ax, pcolormesh)
ani = VideoAnimation(ax.figure, update, **kw_video)
# %%
# Final LAVD
# ^^^^^^^^^^
# %%
# Format LAVD data
lavd = RegularGridDataset.with_array(
coordinates=("lon", "lat"),
datas=dict(
lavd=lavd.T,
lon=x_g,
lat=y_g,
),
centered=True,
)
# %%
# Display final LAVD with py eddy tracker detection.
# Period used for LAVD integration(8 days) is too short for a real use, but choose for example efficiency.
fig, ax, _ = start_ax()
mappable = lavd.display(ax, "lavd", **kw_vorticity)
NetworkObservations.load_file(get_path("Anticyclonic_20160515.nc")).display(
ax, color="k"
)
NetworkObservations.load_file(get_path("Cyclonic_20160515.nc")).display(ax, color="k")
update_axes(ax, mappable)