def eval_simulation(name, site, sim_file=None, plots=False, fix_closure=True,
qc=True, overwrite=False):
"""Main function for evaluating an existing simulation.
Copies simulation data to source directory.
TODO: skip running if cached, for easier page regeneration
:name: name of the model
:site: PALS site name to run the model at
:sim_file: Path to simulation netcdf. Only required if simulation is at a non-standard place.
"""
args = locals()
args_str = '\n'.join([k + ': ' + str(args[k]) for k in sorted(args.keys())])
logger = setup_logger(__name__, 'logs/eval/{m}/{s}/{m}_{s}.log'.format(m=name, s=site))
logger.info("Evaluating model.\nArgs:\n{a}".format(a=args_str))
nc_path = get_sim_nc_path(name, site)
if sim_file is None:
filename = nc_path
else:
filename = sim_file
eval_path = 'source/models/{n}/metrics/{n}_{s}_metrics.csv'.format(n=name, s=site)
if not overwrite:
if os.path.exists(filename) and os.path.exists(eval_path) and \
os.path.getmtime(filename) > os.path.getmtime(eval_path):
logger.warning("Overwriting evaluation file because simulation is newer")
else:
logger.warning("Evaluation file already exists, skipping")
return
try:
sim_data = xr.open_dataset(filename)
except (OSError, RuntimeError) as e:
logger.error("Sim file ({f}) doesn't exist. What are you doing? {e}".format(f=filename, e=e))
return
if sim_file is not None:
logger.warning("Overwriting existing sim!")
sim_data.to_netcdf(nc_path)
flux_data = get_flux_data([site], fix_closure=fix_closure)[site]
evaluate_simulation(sim_data, flux_data, name, site=site, qc=qc)
if plots:
diagnostic_plots(sim_data, flux_data, name, site=site)
sim_data.close()
return
from docopt import docopt
import os
import xarray as xr
import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits import basemap
from mpl_toolkits.axes_grid1 import ImageGrid
from matplotlib.colors import LinearSegmentedColormap
from empirical_lsm.gridded_datasets import get_MODIS_data # , get_GLEAM3a_data, get_MPI_data
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/plot_dataset.log')
def get_range(x):
return np.quantile(x, [0, 100])
def colormap(x0):
"""custom colourmap for map plots"""
if x0 > 0:
x2 = (1 + 2 * x0) / 3.0
x3 = (2 + x0) / 3.0
# blue, white, red, yellow, purple
cdict1 = {
-h, --help Show this screen and exit.
"""
from docopt import docopt
import os
from glob import glob
from empirical_lsm.data import get_sites
from empirical_lsm.checks import check_model_data, check_metrics
from empirical_lsm.offline_simulation import run_model_site_tuples_mp
from empirical_lsm.offline_eval import eval_simulation
from pals_utils.logging import setup_logger
logger = setup_logger(None, 'logs/check_sanity.log')
def main(args):
if args['--sites'] is None:
sites = get_sites('PLUMBER_ext')
else:
try:
sites = get_sites(args['--sites'])
except:
sites = args['--sites'].split(',')
if args['all']:
if args['data']:
models = [os.path.basename(f) for f in glob('model_data/*')]
else:
Usage:
run_model.py run <name> <site> [--no-mp] [--multivariate] [--overwrite] [--no-fix-closure]
Options:
-h, --help Show this screen and exit.
"""
from docopt import docopt
from pals_utils.data import set_config
from empirical_lsm.offline_simulation import run_simulation_mp
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/run_model.log')
set_config(['vars', 'flux'], ['NEE', 'Qle', 'Qh'])
def main(args):
name = args['<name>']
site = args['<site>']
run_simulation_mp(name,
site,
no_mp=args['--no-mp'],
multivariate=args['--multivariate'],
overwrite=args['--overwrite'],
fix_closure=not args['--no-fix-closure'])
def run_simulation(model,
name,
site,
multivariate=False,
overwrite=False,
fix_closure=True):
"""Main function for fitting and running a model.
:model: sklearn-style model or pipeline (regression estimator)
:name: name of the model
:site: PALS site name to run the model at (or 'all', or 'debug')
"""
args = locals()
args_str = '\n'.join(
[k + ': ' + str(args[k]) for k in sorted(args.keys())])
sim_dir = 'model_data/{n}'.format(n=name)
os.makedirs(sim_dir, exist_ok=True)
nc_file = '{d}/{n}_{s}.nc'.format(d=sim_dir, n=name, s=site)
if os.path.isfile(nc_file) and not overwrite:
logger.warning(
"Sim netcdf already exists for {n} at {s}, use --overwrite to re-run."
.format(n=name, s=site))
return
run_logger = setup_logger(
__name__, 'logs/run/{m}/{s}/{m}_{s}.log'.format(m=name, s=site))
run_logger.info("Running model.\nArgs:\n{a}".format(a=args_str))
for i in range(3):
# We attempt to run the model up to 3 times, incase of numerical problems
try:
sim_data = fit_predict(model,
name,
site,
multivariate=multivariate,
fix_closure=fix_closure)
except AssertionError as e:
logger.exception("Model failed: " + str(e))
return
try:
model_sanity_check(sim_data, name, site)
except RuntimeError as e:
logger.warning(str(e))
if i < 2:
logger.warning('Attempting a %s run.' % ['2nd', '3rd'][i])
continue
else:
logger.error(
'Giving up after 3 failed runs. Check your model structres or met data.'
)
sim_data.attrs.update(
{'Warning': 'model failed after 3 attempts, saved anyway'})
else:
# model run successful, presumably
break
if os.path.isfile(nc_file):
logger.warning("Overwriting sim file at {f}".format(f=nc_file))
else:
logger.info("Writing sim file at {f}".format(f=nc_file))
# if site != 'debug':
sim_data.to_netcdf(nc_file)
return
from docopt import docopt
import os
import datetime
import numpy as np
import xarray as xr
import pals_utils.data as pud
from empirical_lsm.data import get_sites, get_data_dir
from empirical_lsm.gridded_datasets import get_dataset_data, get_dataset_freq
from empirical_lsm.models import get_model
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/gridded_benchmarks.log')
def predict_gridded(model, dataset_data, flux_vars, datafreq=None):
"""predict model results for gridded data
:model: scikit-learn style model/pipeline
:data: xarray-style dataset
:returns: xarray-style dataset
"""
# set prediction metadata
prediction = dataset_data[list(dataset_data.coords)]
# Arrays like (var, lon, lat, time)
result = np.full([
inspect_cluster_regression.py <model_path>
inspect_cluster_regression.py (-h | --help | --version)
Options:
-h, --help Show this screen and exit.
--option=<n> Option description [default: 3]
"""
from docopt import docopt
import pickle
import numpy as np
import empirical_lsm
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/inspect_cluster_regression.log')
def get_wrapper_vars(wrapper):
return [v + '_' + l for v, lags in wrapper.var_lags.items() for l in lags]
def get_cluster_regression_model(wrapper):
"""Unwraps a model"""
if isinstance(wrapper, empirical_lsm.clusterregression.ModelByCluster):
return wrapper
else:
return get_cluster_regression_model(wrapper.model)
def get_cluster_regression_centers(model):
import_sim.py benchmark <name> [<site>...]
import_sim.py sim <name> <site> <file>
Options:
-h, --help Show this screen and exit.
"""
from docopt import docopt
import numpy as np
import xarray as xr
from pals_utils.data import get_flux_data
from empirical_lsm.data import get_sites, get_sim_nc_path
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/check_sanity.log')
def fix_benchmark(site_data, name, site):
"""Performs checks on broken benchmarks, and fixes them inplace
:name: benchmark name
:site_data: xarray dataset (will be modified inplace
"""
if name in ['Manabe_Bucket.2', 'Penman_Monteith.1']:
lon = site_data['longitude']
lat = site_data['latitude']
del site_data['longitude'], site_data['latitude'], lon[
'latitude'], lat['longitude']
site_data['longitude'] = lon
Options:
-h, --help Show this screen and exit.
--sites=<sites> Sites to run the models at [default: PLUMBER_ext]
"""
from docopt import docopt
import subprocess
from empirical_lsm.offline_simulation import run_simulation_mp
from empirical_lsm.offline_eval import eval_simulation_mp, main_rst_gen_mp
from empirical_lsm.model_sets import get_model_set
from empirical_lsm.model_search import model_site_index_rst_mp, model_search_index_rst, get_available_models
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/eval_all.log')
def eval_simulation_all(names,
sites,
run=False,
multivariate=True,
evalu=False,
plots=False,
rst=False,
html=False,
rebuild=False,
no_mp=False,
overwrite=False,
fix_closure=True):
Description: Evaluates a model (sim or set of sims) and produces rst output with diagnostics
Usage:
eval_model.py eval <name> <site> [<file>] [--no-mp] [--plot] [--no-fix-closure] [--no-qc]
eval_model.py rst-gen <name> <site> [--no-mp]
Options:
-h, --help Show this screen and exit.
"""
from docopt import docopt
from empirical_lsm.offline_eval import eval_simulation_mp, main_rst_gen_mp
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/eval_model.log')
def main(args):
name = args['<name>']
site = args['<site>']
sim_file = args['<file>']
plots = args['--plot']
if args['eval']:
eval_simulation_mp(name,
site,
sim_file,
plots,
no_mp=args['--no-mp'],
fix_closure=not args['--no-fix-closure'],
model_combos.py (-h | --help | --version)
Options:
-h, --help Show this screen and exit.
--sites=<sites> Sites to run the models at [default: PLUMBER_ext]
"""
from docopt import docopt
from empirical_lsm.model_sets import get_combo_model_names
from empirical_lsm.offline_simulation import run_simulation_mp
from empirical_lsm.offline_eval import eval_simulation_mp
from pals_utils.logging import setup_logger
logger = setup_logger(__name__, 'logs/model_combos.log')
def main(sites, run=False, multivariate=True, evalu=False, plots=False,
no_mp=False, overwrite=False, fix_closure=True):
names = get_combo_model_names()
if args['--run']:
for name in names:
run_simulation_mp(name, sites, no_mp=no_mp, multivariate=multivariate,
overwrite=overwrite, fix_closure=fix_closure)
if args['--eval']:
for name in names:
eval_simulation_mp(name, sites, plots=plots, no_mp=no_mp,
