def _prepare_datasets_for_evaluation(reference, targets, config_data):
""""""
subset = config_data['evaluation'].get('subset', None)
temporal_time_delta = config_data['evaluation'].get(
'temporal_time_delta', None)
spatial_regrid_lats = config_data['evaluation'].get(
'spatial_regrid_lats', None)
spatial_regrid_lons = config_data['evaluation'].get(
'spatial_regrid_lons', None)
# If we have a temporal time delta and it's daily (i.e., 1) we will
# normalize the data as daily data (which means we adjust the start times
# for each bucket of data to be consistent). By default we will normalize
# the data as monthly. Note that this will not break yearly data so it's
# safer to do this no matter what. This keeps us from ending up with 1-off
# errors in the resulting dataset shape post-temporal/spatial adjustments
# that break evaluations.
string_time_delta = 'monthly'
if temporal_time_delta and temporal_time_delta == 1:
string_time_delta = 'daily'
reference = dsp.normalize_dataset_datetimes(reference, string_time_delta)
targets = [
dsp.normalize_dataset_datetimes(t, string_time_delta) for t in targets
]
if subset:
start = dateutil.parser.parse(subset[4])
end = dateutil.parser.parse(subset[5])
bounds = Bounds(subset[0], subset[1], subset[2], subset[3], start, end)
if reference:
reference = dsp.safe_subset(bounds, reference)
if targets:
targets = [dsp.safe_subset(bounds, t) for t in targets]
if temporal_time_delta:
resolution = timedelta(temporal_time_delta)
if reference:
reference = dsp.temporal_rebin(reference, resolution)
if targets:
targets = [dsp.temporal_rebin(t, resolution) for t in targets]
if spatial_regrid_lats and spatial_regrid_lons:
lats = np.arange(spatial_regrid_lats[0], spatial_regrid_lats[1],
spatial_regrid_lats[2])
lons = np.arange(spatial_regrid_lons[0], spatial_regrid_lons[1],
spatial_regrid_lons[2])
if reference:
reference = dsp.spatial_regrid(reference, lats, lons)
if targets:
targets = [dsp.spatial_regrid(t, lats, lons) for t in targets]
return reference, targets
def _prepare_datasets_for_evaluation(reference, targets, config_data):
""""""
subset = config_data['evaluation'].get('subset', None)
temporal_time_delta = config_data['evaluation'].get('temporal_time_delta', None)
spatial_regrid_lats = config_data['evaluation'].get('spatial_regrid_lats', None)
spatial_regrid_lons = config_data['evaluation'].get('spatial_regrid_lons', None)
# If we have a temporal time delta and it's daily (i.e., 1) we will
# normalize the data as daily data (which means we adjust the start times
# for each bucket of data to be consistent). By default we will normalize
# the data as monthly. Note that this will not break yearly data so it's
# safer to do this no matter what. This keeps us from ending up with 1-off
# errors in the resulting dataset shape post-temporal/spatial adjustments
# that break evaluations.
string_time_delta = 'monthly'
if temporal_time_delta and temporal_time_delta == 1:
string_time_delta = 'daily'
reference = dsp.normalize_dataset_datetimes(reference, string_time_delta)
targets = [dsp.normalize_dataset_datetimes(t, string_time_delta) for t in targets]
if subset:
start = dateutil.parser.parse(subset[4])
end = dateutil.parser.parse(subset[5])
bounds = Bounds(subset[0], subset[1], subset[2], subset[3], start, end)
if reference:
reference = dsp.safe_subset(bounds, reference)
if targets:
targets = [dsp.safe_subset(bounds, t) for t in targets]
if temporal_time_delta:
resolution = timedelta(temporal_time_delta)
if reference:
reference = dsp.temporal_rebin(reference, resolution)
if targets:
targets = [dsp.temporal_rebin(t, resolution) for t in targets]
if spatial_regrid_lats and spatial_regrid_lons:
lats = np.arange(spatial_regrid_lats[0], spatial_regrid_lats[1], spatial_regrid_lats[2])
lons = np.arange(spatial_regrid_lons[0], spatial_regrid_lons[1], spatial_regrid_lons[2])
if reference:
reference = dsp.spatial_regrid(reference, lats, lons)
if targets:
targets = [dsp.spatial_regrid(t, lats, lons) for t in targets]
return reference, targets
def test_partial_spatial_overlap(self):
'''Ensure that safe_subset can handle out of bounds spatial values'''
ds = dp.safe_subset(self.target_dataset, self.spatial_out_of_bounds)
spatial_bounds = ds.spatial_boundaries()
self.assertEquals(spatial_bounds[0], -60)
self.assertEquals(spatial_bounds[1], 60)
self.assertEquals(spatial_bounds[2], -170)
self.assertEquals(spatial_bounds[3], 170)
def test_partial_spatial_overlap(self):
'''Ensure that safe_subset can handle out of bounds spatial values'''
ds = dp.safe_subset(self.spatial_out_of_bounds, self.target_dataset)
spatial_bounds = ds.spatial_boundaries()
self.assertEquals(spatial_bounds[0], -60)
self.assertEquals(spatial_bounds[1], 60)
self.assertEquals(spatial_bounds[2], -170)
self.assertEquals(spatial_bounds[3], 170)
def test_partial_temporal_overlap(self):
'''Ensure that safe_subset can handle out of bounds temporal values'''
ds = dp.safe_subset(self.target_dataset, self.temporal_out_of_bounds)
temporal_bounds = ds.time_range()
start = datetime.datetime(2000, 1, 1)
end = datetime.datetime(2009, 12, 1)
self.assertEquals(temporal_bounds[0], start)
self.assertEquals(temporal_bounds[1], end)
def test_partial_temporal_overlap(self):
'''Ensure that safe_subset can handle out of bounds temporal values'''
ds = dp.safe_subset(self.temporal_out_of_bounds, self.target_dataset)
temporal_bounds = ds.time_range()
start = datetime.datetime(2000, 1, 1)
end = datetime.datetime(2009, 12, 1)
self.assertEquals(temporal_bounds[0], start)
self.assertEquals(temporal_bounds[1], end)
def test_entire_bounds_overlap(self):
ds = dp.safe_subset(self.target_dataset, self.everything_out_of_bounds)
spatial_bounds = ds.spatial_boundaries()
temporal_bounds = ds.temporal_boundaries()
start = datetime.datetime(2000, 1, 1)
end = datetime.datetime(2009, 12, 1)
self.assertEquals(spatial_bounds[0], -60)
self.assertEquals(spatial_bounds[1], 60)
self.assertEquals(spatial_bounds[2], -170)
self.assertEquals(spatial_bounds[3], 170)
self.assertEquals(temporal_bounds[0], start)
self.assertEquals(temporal_bounds[1], end)
def test_entire_bounds_overlap(self):
ds = dp.safe_subset(self.everything_out_of_bounds, self.target_dataset)
spatial_bounds = ds.spatial_boundaries()
temporal_bounds = ds.time_range()
start = datetime.datetime(2000, 1, 1)
end = datetime.datetime(2009, 12, 1)
self.assertEquals(spatial_bounds[0], -60)
self.assertEquals(spatial_bounds[1], 60)
self.assertEquals(spatial_bounds[2], -170)
self.assertEquals(spatial_bounds[3], 170)
self.assertEquals(temporal_bounds[0], start)
self.assertEquals(temporal_bounds[1], end)
def _prepare_datasets_for_evaluation(reference, targets, config_data):
""""""
subset = config_data['evaluation'].get('subset', None)
temporal_time_delta = config_data['evaluation'].get('temporal_time_delta', None)
spatial_regrid_lats = config_data['evaluation'].get('spatial_regrid_lats', None)
spatial_regrid_lons = config_data['evaluation'].get('spatial_regrid_lons', None)
if subset:
start = dateutil.parser.parse(subset[4])
end = dateutil.parser.parse(subset[5])
bounds = Bounds(subset[0], subset[1], subset[2], subset[3], start, end)
if reference:
reference = dsp.safe_subset(bounds, reference)
if targets:
targets = [dsp.safe_subset(bounds, t) for t in targets]
if temporal_time_delta:
resolution = timedelta(temporal_time_delta)
if reference:
reference = dsp.temporal_rebin(reference, resolution)
if targets:
targets = [dsp.temporal_rebin(t, resolution) for t in targets]
if spatial_regrid_lats and spatial_regrid_lons:
lats = np.arange(spatial_regrid_lats[0], spatial_regrid_lats[1], spatial_regrid_lats[2])
lons = np.arange(spatial_regrid_lons[0], spatial_regrid_lons[1], spatial_regrid_lons[2])
if reference:
reference = dsp.spatial_regrid(reference, lats, lons)
if targets:
targets = [dsp.spatial_regrid(t, lats, lons) for t in targets]
return reference, targets
def run_evaluation():
''' Run an OCW Evaluation.
*run_evaluation* expects the Evaluation parameters to be POSTed in
the following format.
.. sourcecode:: javascript
{
reference_dataset: {
// Id that tells us how we need to load this dataset.
'data_source_id': 1 == local, 2 == rcmed,
// Dict of data_source specific identifying information.
//
// if data_source_id == 1 == local:
// {
// 'id': The path to the local file on the server for loading.
// 'var_name': The variable data to pull from the file.
// 'lat_name': The latitude variable name.
// 'lon_name': The longitude variable name.
// 'time_name': The time variable name
// 'name': Optional dataset name
// }
//
// if data_source_id == 2 == rcmed:
// {
// 'dataset_id': The dataset id to grab from RCMED.
// 'parameter_id': The variable id value used by RCMED.
// 'name': Optional dataset name
// }
'dataset_info': {..}
},
// The list of target datasets to use in the Evaluation. The data
// format for the dataset objects should be the same as the
// reference_dataset above.
'target_datasets': [{...}, {...}, ...],
// All the datasets are re-binned to the reference dataset
// before being added to an experiment. This step (in degrees)
// is used when re-binning both the reference and target datasets.
'spatial_rebin_lat_step': The lat degree step. Integer > 0,
// Same as above, but for lon
'spatial_rebin_lon_step': The lon degree step. Integer > 0,
// The temporal resolution to use when doing a temporal re-bin
// This is a timedelta of days to use so daily == 1, monthly is
// (1, 31], annual/yearly is (31, 366], and full is anything > 366.
'temporal_resolution': Integer in range(1, 999),
// A list of the metric class names to use in the evaluation. The
// names must match the class name exactly.
'metrics': [Bias, TemporalStdDev, ...]
// The bounding values used in the Evaluation. Note that lat values
// should range from -180 to 180 and lon values from -90 to 90.
'start_time': start time value in the format '%Y-%m-%d %H:%M:%S',
'end_time': end time value in the format '%Y-%m-%d %H:%M:%S',
'lat_min': The minimum latitude value,
'lat_max': The maximum latitude value,
'lon_min': The minimum longitude value,
'lon_max': The maximum longitude value,
// NOTE: At the moment, subregion support is fairly minimal. This
// will be addressed in the future. Ideally, the user should be able
// to load a file that they have locally. That would change the
// format that this data is passed.
'subregion_information': Path to a subregion file on the server.
}
'''
# TODO: validate input parameters and return an error if not valid
eval_time_stamp = datetime.now().strftime('%Y-%m-%d_%H-%M-%S')
data = request.json
eval_bounds = {
'start_time': datetime.strptime(data['start_time'], '%Y-%m-%d %H:%M:%S'),
'end_time': datetime.strptime(data['end_time'], '%Y-%m-%d %H:%M:%S'),
'lat_min': float(data['lat_min']),
'lat_max': float(data['lat_max']),
'lon_min': float(data['lon_min']),
'lon_max': float(data['lon_max'])
}
# Load all the datasets
ref_dataset = _process_dataset_object(data['reference_dataset'], eval_bounds)
target_datasets = [_process_dataset_object(obj, eval_bounds)
for obj
in data['target_datasets']]
# Normalize the dataset time values so they break on consistent days of the
# month or time of the day, depending on how they will be rebinned.
resolution = data['temporal_resolution']
time_delta = timedelta(days=resolution)
time_step = 'daily' if resolution == 1 else 'monthly'
ref_dataset = dsp.normalize_dataset_datetimes(ref_dataset, time_step)
target_datasets = [dsp.normalize_dataset_datetimes(ds, time_step)
for ds in target_datasets]
# Subset the datasets
start = eval_bounds['start_time']
end = eval_bounds['end_time']
# Normalize all the values to the first of the month if we're not
# dealing with daily data. This will ensure that a valid subregion
# isn't considered out of bounds do to a dataset's time values
# being shifted to the first of the month.
if time_step != 'daily':
if start.day != 1:
day_offset = start.day - 1
start -= timedelta(days=day_offset)
if end.day != 1:
day_offset = end.day - 1
end -= timedelta(days=day_offset)
subset = Bounds(eval_bounds['lat_min'],
eval_bounds['lat_max'],
eval_bounds['lon_min'],
eval_bounds['lon_max'],
start,
end)
ref_dataset = dsp.safe_subset(subset, ref_dataset)
target_datasets = [dsp.safe_subset(subset, ds)
for ds
in target_datasets]
# Do temporal re-bin based off of passed resolution
ref_dataset = dsp.temporal_rebin(ref_dataset, time_delta)
target_datasets = [dsp.temporal_rebin(ds, time_delta)
for ds
in target_datasets]
# Do spatial re=bin based off of reference dataset + lat/lon steps
lat_step = data['spatial_rebin_lat_step']
lon_step = data['spatial_rebin_lon_step']
lat_bins, lon_bins = _calculate_new_latlon_bins(eval_bounds,
lat_step,
lon_step)
ref_dataset = dsp.spatial_regrid(ref_dataset, lat_bins, lon_bins)
target_datasets = [dsp.spatial_regrid(ds, lat_bins, lon_bins)
for ds
in target_datasets]
# Load metrics
loaded_metrics = _load_metrics(data['metrics'])
# Prime evaluation object with data
evaluation = Evaluation(ref_dataset, target_datasets, loaded_metrics)
# Run evaluation
evaluation.run()
# Plot
_generate_evaluation_plots(evaluation, lat_bins, lon_bins, eval_time_stamp)
return json.dumps({'eval_work_dir': eval_time_stamp})
def run_evaluation():
''' Run an OCW Evaluation.
*run_evaluation* expects the Evaluation parameters to be POSTed in
the following format.
.. sourcecode:: javascript
{
reference_dataset: {
// Id that tells us how we need to load this dataset.
'data_source_id': 1 == local, 2 == rcmed,
// Dict of data_source specific identifying information.
//
// if data_source_id == 1 == local:
// {
// 'id': The path to the local file on the server for loading.
// 'var_name': The variable data to pull from the file.
// 'lat_name': The latitude variable name.
// 'lon_name': The longitude variable name.
// 'time_name': The time variable name
// 'name': Optional dataset name
// }
//
// if data_source_id == 2 == rcmed:
// {
// 'dataset_id': The dataset id to grab from RCMED.
// 'parameter_id': The variable id value used by RCMED.
// 'name': Optional dataset name
// }
'dataset_info': {..}
},
// The list of target datasets to use in the Evaluation. The data
// format for the dataset objects should be the same as the
// reference_dataset above.
'target_datasets': [{...}, {...}, ...],
// All the datasets are re-binned to the reference dataset
// before being added to an experiment. This step (in degrees)
// is used when re-binning both the reference and target datasets.
'spatial_rebin_lat_step': The lat degree step. Integer > 0,
// Same as above, but for lon
'spatial_rebin_lon_step': The lon degree step. Integer > 0,
// The temporal resolution to use when doing a temporal re-bin
// This is a timedelta of days to use so daily == 1, monthly is
// (1, 31], annual/yearly is (31, 366], and full is anything > 366.
'temporal_resolution': Integer in range(1, 999),
// A list of the metric class names to use in the evaluation. The
// names must match the class name exactly.
'metrics': [Bias, TemporalStdDev, ...]
// The bounding values used in the Evaluation. Note that lat values
// should range from -180 to 180 and lon values from -90 to 90.
'start_time': start time value in the format '%Y-%m-%d %H:%M:%S',
'end_time': end time value in the format '%Y-%m-%d %H:%M:%S',
'lat_min': The minimum latitude value,
'lat_max': The maximum latitude value,
'lon_min': The minimum longitude value,
'lon_max': The maximum longitude value,
// NOTE: At the moment, subregion support is fairly minimal. This
// will be addressed in the future. Ideally, the user should be able
// to load a file that they have locally. That would change the
// format that this data is passed.
'subregion_information': Path to a subregion file on the server.
}
'''
# TODO: validate input parameters and return an error if not valid
eval_time_stamp = datetime.now().strftime('%Y-%m-%d_%H-%M-%S')
data = request.json
eval_bounds = {
'start_time': datetime.strptime(data['start_time'], '%Y-%m-%d %H:%M:%S'),
'end_time': datetime.strptime(data['end_time'], '%Y-%m-%d %H:%M:%S'),
'lat_min': float(data['lat_min']),
'lat_max': float(data['lat_max']),
'lon_min': float(data['lon_min']),
'lon_max': float(data['lon_max'])
}
# Load all the datasets
ref_dataset = _process_dataset_object(data['reference_dataset'], eval_bounds)
target_datasets = [_process_dataset_object(obj, eval_bounds)
for obj
in data['target_datasets']]
# Normalize the dataset time values so they break on consistent days of the
# month or time of the day, depending on how they will be rebinned.
resolution = data['temporal_resolution']
time_delta = timedelta(days=resolution)
time_step = 'daily' if resolution == 1 else 'monthly'
ref_dataset = dsp.normalize_dataset_datetimes(ref_dataset, time_step)
target_datasets = [dsp.normalize_dataset_datetimes(ds, time_step)
for ds in target_datasets]
# Subset the datasets
start = eval_bounds['start_time']
end = eval_bounds['end_time']
# Normalize all the values to the first of the month if we're not
# dealing with daily data. This will ensure that a valid subregion
# isn't considered out of bounds do to a dataset's time values
# being shifted to the first of the month.
if time_step != 'daily':
if start.day != 1:
day_offset = start.day - 1
start -= timedelta(days=day_offset)
if end.day != 1:
day_offset = end.day - 1
end -= timedelta(days=day_offset)
subset = Bounds(eval_bounds['lat_min'],
eval_bounds['lat_max'],
eval_bounds['lon_min'],
eval_bounds['lon_max'],
start,
end)
ref_dataset = dsp.safe_subset(ref_dataset, subset)
target_datasets = [dsp.safe_subset(ds, subset)
for ds
in target_datasets]
# Do temporal re-bin based off of passed resolution
ref_dataset = dsp.temporal_rebin(ref_dataset, time_delta)
target_datasets = [dsp.temporal_rebin(ds, time_delta)
for ds
in target_datasets]
# Do spatial re=bin based off of reference dataset + lat/lon steps
lat_step = data['spatial_rebin_lat_step']
lon_step = data['spatial_rebin_lon_step']
lat_bins, lon_bins = _calculate_new_latlon_bins(eval_bounds,
lat_step,
lon_step)
ref_dataset = dsp.spatial_regrid(ref_dataset, lat_bins, lon_bins)
target_datasets = [dsp.spatial_regrid(ds, lat_bins, lon_bins)
for ds
in target_datasets]
# Load metrics
loaded_metrics = _load_metrics(data['metrics'])
# Prime evaluation object with data
evaluation = Evaluation(ref_dataset, target_datasets, loaded_metrics)
# Run evaluation
evaluation.run()
# Plot
_generate_evaluation_plots(evaluation, lat_bins, lon_bins, eval_time_stamp)
return json.dumps({'eval_work_dir': eval_time_stamp})
