def main(config_yaml):
with open(config_yaml, 'r') as fp:
config_str = jinja2.Template(fp.read()).render()
config = yaml.load(config_str, Loader=yaml.Loader)
# Get the classes associated with each
for name in config['run']:
try:
config['cls'] = tc.Tomographer._find_subclass(name)
except KeyError:
raise ValueError(f"Tomographer {name} is not defined")
# Decide if anyone needs the colors calculating and/or errors loading
anyone_wants_colors = False
anyone_wants_errors = False
for run in config['run'].values():
for version in run.values():
if version.get('errors'):
anyone_wants_errors = True
if version.get('colors'):
anyone_wants_colors = True
bands = config['bands']
training_data = tc.load_data(
config['training_file'],
bands,
errors=anyone_wants_errors,
colors=anyone_wants_colors
)
validation_data = tc.load_data(
config['validation_file'],
bands,
errors=anyone_wants_errors,
colors=anyone_wants_colors
)
training_z = tc.load_redshift(config['training_file'])
validation_z = tc.load_redshift(config['validation_file'])
if (('metrics_impl' not in config) or
(config['metrics_impl'] == 'firecrown')):
metrics_fn = tc.compute_scores
elif config['metrics_impl'] == 'jax-cosmo':
metrics_fn = tc.jc_compute_scores
else:
raise ValueError('Unknown metrics_impl value')
with open(config['output_file'],'w') as output_file:
for classifier_name, runs in config['run'].items():
for run, settings in runs.items():
scores = run_one(classifier_name, bands, settings,
training_data, training_z, validation_data, validation_z,
config['metrics'], metrics_fn)
output_file.write (f"{classifier_name} {run} {settings} {scores} \n")
def load_data(fname, bands):
data = tc.load_magnitudes_and_colors(fname, bands)
Nbands = len(bands)
print('max mag before:', data[:, :Nbands].max(axis=0), 'nobj:',
data.shape[0])
z = tc.load_redshift(fname)
data, z = tc.add_noise_snr_cut(data, z, bands)
print('max mag after:', data[:, :Nbands].max(axis=0), 'nobj:',
data.shape[0])
return data, z