def run(self):
self.out.set_fraction(0.01)
points = self.points_store['points']
training_dates = self.training_data_store.keys()
training_array = np.zeros((len(points), len(training_dates)))
# build array of training data
for date_i in range(len(training_dates)):
date_str = training_dates[date_i]
print date_str
data = self.training_data_store[date_str]
data_year, data_month, data_day = [int(s) for s in date_str.split('-')]
data_midnight = datetime.datetime(data_year, data_month, data_day, 0, 0)
time_offset = ppas.datetime_to_seconds(self.exp_midnight) - ppas.datetime_to_seconds(data_midnight)
for p_i in range(len(points)):
p = points[p_i]
t = p[3] - time_offset
val = ppas.roms.get_value(data, p[0], p[1], p[2], t, self.varname, interp='linear')
training_array[p_i,date_i] = val
self.out.set_fraction(0.5)
# filter out bad data
if self.varname == 'salt':
varmin, varmax = 30., 40.
elif self.varname == 'temp':
varmin, varmax = 10., 100.
else:
dieee_unknown_var_type___
good_columns = ((training_array > varmin) & ( training_array < varmax)).all(axis=0)
self.out.set_fraction(0.7)
# calculate covariance matrix
kmat = np.cov(training_array[:,good_columns])
self.save_store['kmat'] = kmat
self.out.set_fraction(1.0)
def run(self):
self.out.set_fraction(0.01)
points = self.points_store['points']
training_dates = self.training_data_store.keys()
training_array = np.zeros((len(points), len(training_dates)))
# build array of training data
for date_i in range(len(training_dates)):
date_str = training_dates[date_i]
print date_str
data = self.training_data_store[date_str]
data_year, data_month, data_day = [
int(s) for s in date_str.split('-')
]
data_midnight = datetime.datetime(data_year, data_month, data_day,
0, 0)
time_offset = ppas.datetime_to_seconds(
self.exp_midnight) - ppas.datetime_to_seconds(data_midnight)
for p_i in range(len(points)):
p = points[p_i]
t = p[3] - time_offset
val = ppas.roms.get_value(data,
p[0],
p[1],
p[2],
t,
self.varname,
interp='linear')
training_array[p_i, date_i] = val
self.out.set_fraction(0.5)
# filter out bad data
if self.varname == 'salt':
varmin, varmax = 30., 40.
elif self.varname == 'temp':
varmin, varmax = 10., 100.
else:
dieee_unknown_var_type___
good_columns = ((training_array > varmin) &
(training_array < varmax)).all(axis=0)
self.out.set_fraction(0.7)
# calculate covariance matrix
kmat = np.cov(training_array[:, good_columns])
self.save_store['kmat'] = kmat
self.out.set_fraction(1.0)
def calc_emse(run_dir):
settings = pplan.PPlanSettings(run_dir)
store = ppas.Store(settings.data_dir)
P = store['P']
G = store['G']
objective = ppas.objectives.EMSEObjective(store['kmat'], settings.planner_properties['sigma_n'])
start_t = ppas.datetime_to_seconds(settings.roi_properties['starttime'])
samples = ppas.graph.path_samples(store['P'], start_t)
print '%s: EMSE:' % run_dir, objective.f(samples)
def calc_emp_cov(run_dir, args):
print '%s: calculating empirical covariance matrix' % run_dir
settings = pplan.PPlanSettings(run_dir)
store = ppas.Store(settings.data_dir)
training_store = ppas.Store(os.path.join(settings.data_dir, 'training_data'))
points = store['points']
qoi = settings.roi_properties['qoi']
training_dates = sorted(training_store.keys())
training_array = np.zeros((len(points), len(training_dates)))
exp_start = settings.roi_properties['starttime']
exp_midnight = datetime.datetime(exp_start.year, exp_start.month, exp_start.day, 0, 0)
# build array of training data
for date_i in range(len(training_dates)):
date_str = training_dates[date_i]
data = training_store[date_str]
data_year, data_month, data_day = [int(s) for s in date_str.split('-')]
data_midnight = datetime.datetime(data_year, data_month, data_day, 0, 0)
if exp_start.hour < 13:
exp_midnight += datetime.timedelta(days=1)
time_offset = ppas.datetime_to_seconds(exp_midnight) - ppas.datetime_to_seconds(data_midnight)
for p_i in range(len(points)):
p = points[p_i]
t = p[3] - time_offset
val = ppas.roms.get_value(data, p[0], p[1], p[2], t, qoi, interp='linear')
training_array[p_i,date_i] = val
# filter out bad data
if qoi == 'salt':
varmin, varmax = 30., 40.
elif qoi == 'temp':
varmin, varmax = 0., 100.
else:
dieee_unknown_var_type___
good_columns = ((training_array > varmin) & ( training_array < varmax)).all(axis=0)
good_training_data = training_array[:,good_columns]
store['good_training_data'] = good_training_data
# calculate covariance matrix
kmat = np.cov(good_training_data)
store['kmat'] = kmat
def parse_time(date_str, time_str, timezone=pytz.utc):
''' Parse the ecomapper time format
date_str is in format mm/dd/yyyy
time_str is format hh:mm:ss.ss
Returns ppas time as defined in ppas/util.py
'''
month_str, day_str, year_str = date_str.split('/')
hour_str, min_str, float_sec_str = time_str.split(':')
float_sec = float(float_sec_str)
dt = datetime.datetime(int(year_str), int(month_str), int(day_str),
int(hour_str), int(min_str), int(float_sec), int((float_sec%1)*1e6))
ldt = timezone.localize(dt)
return ppas.datetime_to_seconds(ldt)
roslib.load_manifest('ppas')
import datetime
import numpy as np
import ppas
lat0 = 33.4967
lon0 = -118.72
lat1 = 33.58
lon1 = -118.52
depth0 = 30.
depth1 = 50.
year, month, day = 2011, 7, 26
# get 3 hours of data starting at 15:00 GMT
time0 = ppas.datetime_to_seconds(datetime.datetime(year, month, day, 15))
time1 = time0 + 3. * 3600.
print 'Connecting to ROMS server'
dataset = ppas.roms.open_dataset(datetime.date(year, month, day))
print 'Downloading data'
data = ppas.roms.get_data(dataset, lat0, lat1, lon0, lon1, depth0, depth1,
time0, time1)
# print somve values
lat = (lat0 + lat1) / 2.0
lon = (lon0 + lon1) / 2.0
depth = (depth0 + depth1) / 2.0
for t in np.linspace(time0, time1):
v = ppas.roms.get_value(data, lat, lon, depth, t, 'temp', interp='linear')
import roslib
roslib.load_manifest('ppas')
import time, datetime
import ppas
# get the current POSIX time
t = time.time()
# convert to UTC datetime
tm = time.gmtime(t)
dt = datetime.datetime(tm.tm_year, tm.tm_mon, tm.tm_mday, tm.tm_hour,
tm.tm_min, tm.tm_sec)
# convert back to POSIX time using our own function
t_ppas = ppas.datetime_to_seconds(dt)
# error should be less than one second (time.gmtime rounds to the nearest second)
print 'Error:', t - t_ppas
def makegraph_edgetime_equilateral(roi_properties, graph_properties):
'''
Construct a graph where all edges have equal lengths
'''
spatial_points = []
nrows, ncols = graph_properties['shape']
edge_distance = graph_properties['edge_distance']
row_spacing = ppas.roms.meters_to_degrees(edge_distance, 0.0)[0]
col_spacing = ppas.roms.meters_to_degrees(
0.0, np.cos(np.pi/6.)*edge_distance)[1]
for col_i in range(ncols):
lon = col_i * col_spacing + roi_properties['lon0']
if col_i % 2 == 1:
offset = 0.5 * row_spacing
nrows_thiscol = nrows - 1
else:
offset = 0.0
nrows_thiscol = nrows
for row_i in range(nrows_thiscol):
lat = offset + row_spacing * row_i + roi_properties['lat0']
spatial_points.append(np.array((lat, lon)))
spatial_points = np.array(spatial_points)
# make the graph
nodes = range(len(spatial_points))
G = ppas.graph.Graph(nodes, 'discrete_time')
G.node_points = spatial_points
starttime = ppas.datetime_to_seconds(roi_properties['starttime'])
points = []
for v_i in nodes:
sp_i = spatial_points[v_i]
for v_j in nodes:
sp_j = spatial_points[v_j]
meters_delta = ppas.roms.degrees_to_meters(
sp_i[0] - sp_j[0], sp_i[1] - sp_j[1])
distance = linalg.norm(meters_delta)
if distance <= edge_distance * 1.01 and v_i != v_j: # fudge factor...
length_dict = {}
sample_dict = {}
for t in graph_properties['time_list']:
length_dict[t] = graph_properties['edge_len']
samples = set()
ppe = graph_properties['ppe']
# technically each sample should be at a different time, but
# then we would end up with more points (due to points being
# at different times depending on which direction the edge
# is taversed)
t_s = 0.5 * t + 0.5 * (t + graph_properties['edge_len'])
for r in np.linspace(0., 1., ppe):
sp = (1 - r)*sp_i + r*sp_j
depth = roi_properties['depth']
p = np.array((sp[0], sp[1], depth, t_s))
p_ii = add_point(points, p)
samples.add(p_ii)
sample_dict[t] = samples
e = ppas.graph.DiscreteTimeEdge(v_i, v_j, length_dict, sample_dict)
G.add_edge(e)
return G, np.array(points)
roslib.load_manifest('ppas')
import datetime
import numpy as np
import ppas
lat0 = 33.4967
lon0 = -118.72
lat1 = 33.58
lon1 = -118.52
depth0 = 30.
depth1 = 50.
year, month, day = 2011, 7, 26
# get 3 hours of data starting at 15:00 GMT
time0 = ppas.datetime_to_seconds(datetime.datetime(year, month, day, 15))
time1 = time0 + 3.*3600.
print 'Connecting to ROMS server'
dataset = ppas.roms.open_dataset(datetime.date(year, month, day))
print 'Downloading data'
data = ppas.roms.get_data(dataset, lat0, lat1, lon0, lon1, depth0, depth1, time0, time1)
# print somve values
lat = (lat0 + lat1)/2.0
lon = (lon0 + lon1)/2.0
depth = (depth0 + depth1)/2.0
for t in np.linspace(time0, time1):
v = ppas.roms.get_value(data, lat, lon, depth, t, 'temp', interp='linear')
print 'Temperature at (%.3f, %.3f, %.1f, %.7f): %f' % (lat, lon, depth, t, v)
def makegraph_edgetime_equilateral(roi_properties, graph_properties):
'''
Construct a graph where all edges have equal lengths
'''
spatial_points = []
nrows, ncols = graph_properties['shape']
edge_distance = graph_properties['edge_distance']
row_spacing = ppas.roms.meters_to_degrees(edge_distance, 0.0)[0]
col_spacing = ppas.roms.meters_to_degrees(
0.0,
np.cos(np.pi / 6.) * edge_distance)[1]
for col_i in range(ncols):
lon = col_i * col_spacing + roi_properties['lon0']
if col_i % 2 == 1:
offset = 0.5 * row_spacing
nrows_thiscol = nrows - 1
else:
offset = 0.0
nrows_thiscol = nrows
for row_i in range(nrows_thiscol):
lat = offset + row_spacing * row_i + roi_properties['lat0']
spatial_points.append(np.array((lat, lon)))
spatial_points = np.array(spatial_points)
# make the graph
nodes = range(len(spatial_points))
G = ppas.graph.Graph(nodes, 'discrete_time')
G.node_points = spatial_points
starttime = ppas.datetime_to_seconds(roi_properties['starttime'])
points = []
for v_i in nodes:
sp_i = spatial_points[v_i]
for v_j in nodes:
sp_j = spatial_points[v_j]
meters_delta = ppas.roms.degrees_to_meters(sp_i[0] - sp_j[0],
sp_i[1] - sp_j[1])
distance = linalg.norm(meters_delta)
if distance <= edge_distance * 1.01 and v_i != v_j: # fudge factor...
length_dict = {}
sample_dict = {}
for t in graph_properties['time_list']:
length_dict[t] = graph_properties['edge_len']
samples = set()
ppe = graph_properties['ppe']
# technically each sample should be at a different time, but
# then we would end up with more points (due to points being
# at different times depending on which direction the edge
# is taversed)
t_s = 0.5 * t + 0.5 * (t + graph_properties['edge_len'])
for r in np.linspace(0., 1., ppe):
sp = (1 - r) * sp_i + r * sp_j
depth = roi_properties['depth']
p = np.array((sp[0], sp[1], depth, t_s))
p_ii = add_point(points, p)
samples.add(p_ii)
sample_dict[t] = samples
e = ppas.graph.DiscreteTimeEdge(v_i, v_j, length_dict,
sample_dict)
G.add_edge(e)
return G, np.array(points)