def parade(p: dict, x: [float], y: float):
""" Process an observation y
:param p: state - supply empty dict on first call
:param y: incoming observation
:param x: term structure of predictions out k steps ahead, made after y received
returns: mean, std, of model residuals and the posterior state s'
"""
# Initialize
if not p:
k = len(x)
p = {
'predictions': [[] for _ in range(k)], # Holds the cavalcade
'moments': [var_init() for _ in range(k)]
} # Could use kurtosis_init here for more moment
else:
assert len(x) == len(p['predictions']) # 'k' is immutable
assessable = p['predictions'].pop(0)
if assessable:
for j, xi in assessable:
p['moments'][j] = var_update(p['moments'][j], y - xi)
p['predictions'].append(list())
for j, xj in enumerate(x):
p['predictions'][j].append((j, xj))
return parade_mean(p), parade_std(p), p
def parade(p:dict, x:Union[List[SupportsFloat],None], y:Union[SupportsFloat,None]):
""" Process an observation y
:param p: state - supply empty dict on first call
:param y: incoming observation
:param x: term structure of predictions out k steps ahead, made after y received
returns: mean, std, of model residuals and the posterior state s'
A special convention allows the caller to reset the empirical moments. Pass x=None and y=None
"""
# Initialize
if not p:
k = len(x)
p = {'predictions': [[] for _ in range(k)], # Holds the cavalcade
'moments': [var_init() for _ in range(k)]} # Could use kurtosis_init here for more moment
else:
assert len(x) == len(p['predictions']) # 'k' is immutable
if x is None and y is None:
# "reset" the running moments but keep prediction parade
p_mean, p_std = parade_mean(p), parade_std(p)
p['moments'] = [var_init() for _ in range(k)]
return p_mean, p_std, p
else:
assessable = p['predictions'].pop(0)
if assessable:
for j,xi in assessable:
p['moments'][j] = var_update(p['moments'][j], y - xi)
p['predictions'].append(list())
for j, xj in enumerate(x):
p['predictions'][j].append((j, xj))
return parade_mean(p), parade_std(p), p
n_train) + '_k_' + str(k) + '_multiple_' + str(
round(multiple, 1))
try:
os.mkdir(jpg_path)
except FileExistsError:
pass
jpg_file = jpg_path + os.path.sep + 'rel=' + str(
rel_error) + '_name=' + name.replace('.json', '.jpg')
plt.savefig(jpg_file)
plt.close()
print(jpg_file)
return err, lv_err, jpg_path
if __name__ == '__main__':
from momentum import var_init, var_update
import json
lv = var_init()
pr = var_init()
params = {'multiple': 1.5, 'n_recent': 5, 'n_train': 500, 'k': 20}
while True:
err, lv_err, jpg_path = plot_next_optionated_forecast(**params)
report = params.copy()
report['simple_average'] = var_update(lv, lv_err)
report['prophet'] = var_update(pr, err)
pprint(report)
report_file = jpg_path + '_summary.json'
with open(report_file, 'wt') as f:
json.dump(report, f)