args, unknown = argparser.parse_known_args()
p = Config(args.config_path)
models = p.model_list[:]
lm_formula = p.models['LMnoS_noRE']['formula']
preds = get_preds(lm_formula)
preds = list(preds)
preds.sort(reverse=True, key=lambda x: len(x))
n_pred = len(preds)
perms = list(itertools.product(range(0, 4), repeat=n_pred))
forms = permute_spillover(lm_formula, preds, perms)
dtsr_formula_list = [Formula(p.models[m]['formula']) for m in p.model_list if m.startswith('DTSR')]
dtsr_formula_name_list = [m for m in p.model_list if m.startswith('DTSR')]
X, y = read_data(p.X_train, p.y_train, p.series_ids, categorical_columns=list(set(p.series_ids + [v for x in dtsr_formula_list for v in x.rangf])))
X, y, select, X_2d_predictor_names, X_2d_predictors = preprocess_data(X, y, p, dtsr_formula_list, compute_history=False)
from dtsr.baselines import py2ri
X['splitID'] = compute_splitID(X, p.split_ids)
part = compute_partition(X, p.modulus, 3)
part_select = part[0]
X_baseline = X[part_select]
X_baseline = X_baseline.reset_index(drop=True)[select]
n_train_sample = len(y)
sys.stderr.write('\nNumber of training samples: %d\n' %n_train_sample)
p = Config(args.config_path)
if len(args.models) > 0:
models = args.models
else:
models = p.model_list[:]
run_baseline = False
run_dtsr = False
for m in models:
if not run_baseline and m.startswith('LM') or m.startswith('GAM'):
run_baseline = True
elif not run_dtsr and m.startswith('DTSR'):
run_dtsr = True
dtsr_formula_list = [
Formula(p.models[m]['formula']) for m in p.model_list
if m.startswith('DTSR')
]
dtsr_formula_name_list = [m for m in p.model_list if m.startswith('DTSR')]
if args.partition == 'train':
X, y = read_data(
p.X_train,
p.y_train,
p.series_ids,
categorical_columns=list(
set(p.split_ids + p.series_ids +
[v for x in dtsr_formula_list for v in x.rangf])))
elif args.partition == 'dev':
X, y = read_data(
p.X_dev,
p.y_dev,
default_irf_str = impulse_to_irf['default']
else:
default_irf_str = 'ShiftedGammaShapeGT1(alpha=2, beta=5, delta=-0.2, ran=T)'
impulse_to_transform = {}
if 'impulse_to_transform' in config:
for key in config['impulse_to_transform']:
impulse_to_transform[key] = config['impulse_to_transform'][key]
baseline_found = False
for name in config:
if name.startswith('model_DTSR') and name.endswith('_BASELINE'):
baseline_found = True
new_name = name[:-9]
model_template = config[name]
f = Formula(model_template['formula'])
for effect in main_effects:
irf_str = impulse_to_irf.get(effect, default_irf_str)
effect_name = effect
transform = impulse_to_transform.get(effect, ['s'])
if transform == 'None':
transform = []
elif isinstance(transform, basestring):
transform = transform.strip().split()
for t in reversed(transform):
effect_name = t + '(' + effect_name + ')'
f.insert_impulses(effect_name, irf_str, rangf=['subject'])
config[new_name] = dict(config[name]).copy()
default_irf_str = impulse_to_irf['default']
else:
default_irf_str = 'ShiftedGammaShapeGT1(alpha=2, beta=5, delta=-0.2, ran=T)'
impulse_to_transform = {}
if 'impulse_to_transform' in config:
for key in config['impulse_to_transform']:
impulse_to_transform[key] = config['impulse_to_transform'][key]
baseline_found = False
for name in config:
if name.startswith('model_DTSR') and name.endswith('_BASELINE'):
baseline_found = True
new_name = name[:-9]
model_template = config[name]
f = Formula(model_template['formula'])
for effect in main_effects:
irf_str = impulse_to_irf.get(effect, default_irf_str)
effect_name = effect
transform = impulse_to_transform.get(effect, ['s'])
if transform == 'None':
transform = []
elif isinstance(transform, basestring):
transform = transform.strip().split()
for t in reversed(transform):
effect_name = t + '(' + effect_name + ')'
f.insert_impulses(effect_name, irf_str, rangf=['subject'])
config[new_name] = dict(config[name]).copy()
impulse_to_transform = {}
if 'impulse_to_transform' in config:
for key in config['impulse_to_transform']:
impulse_to_transform[key] = config['impulse_to_transform'][key]
newpred_rangf = ['subject']
if 'newpred_rangf' in config:
newpred_rangf = config['newpred_rangf']['rangf'].strip().split()
baseline_found = False
for name in config:
if name.startswith('model_DTSR') and name.endswith('_BASELINE'):
baseline_found = True
new_name = name[:-9]
model_template = config[name]
f = Formula(model_template['formula'])
for effect in main_effects:
if effect.startswith('~'):
ablate = True
effect = effect[1:]
else:
ablate = False
irf_str = impulse_to_irf.get(effect, default_irf_str)
effect_name = effect
transform = impulse_to_transform.get(effect, ['s'])
if transform == 'None':
transform = []
elif isinstance(transform, str):
transform = transform.strip().split()
impulse_to_transform = {}
if 'impulse_to_transform' in config:
for key in config['impulse_to_transform']:
impulse_to_transform[key] = config['impulse_to_transform'][key]
newpred_rangf = ['subject']
if 'newpred_rangf' in config:
newpred_rangf = config['newpred_rangf']['rangf'].strip().split()
baseline_found = False
for name in config:
if name.startswith('model_DTSR') and name.endswith('_BASELINE'):
baseline_found = True
new_name = name[:-9]
model_template = config[name]
f = Formula(model_template['formula'])
for effect in main_effects:
if effect.startswith('~'):
ablate = True
effect = effect[1:]
else:
ablate = False
irf_str = impulse_to_irf.get(effect, default_irf_str)
effect_name = effect
transform = impulse_to_transform.get(effect, ['s'])
if transform == 'None':
transform = []
elif isinstance(transform, str):
transform = transform.strip().split()
import sys
import configparser
sys.path.append(sys.argv[1])
from dtsr.formula import Formula
config = configparser.ConfigParser()
config.optionxform = str
config.readfp(sys.stdin)
cols = ['word', 'time', 'subject']
cols += config['data']['series_ids'].strip().split()
if 'split_ids' in config['data']:
cols += config['data']['split_ids'].strip().split()
for m in [x for x in config if x.startswith('model_DTSR')]:
formula = Formula(config[m]['formula'])
cols.append(formula.dv_term.id)
for impulse in formula.t.impulses():
cols.append(impulse.id)
cols = sorted(list(set(cols)))
print(' '.join(cols))