Skip to content

Commit

Permalink
corrected optimizatin strategy for multi-targets. Corrected omegas
Browse files Browse the repository at this point in the history
  • Loading branch information
@nepslor
nepslor committed Apr 10, 2024
1 parent 28c600c commit 60b597f
Show file tree
Hide file tree
Showing 2 changed files with 20 additions and 16 deletions.
30 changes: 17 additions & 13 deletions pyforecaster/forecasting_models/holtwinters.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,10 +19,11 @@ def hankel(x, n, generate_me=None):
h.append(x[w + g].reshape(-1, 1))
return np.hstack(h)

def fit_sample(pars_dict,model_class, model_init_kwargs, x, targets_names=None, return_model=False):
def fit_sample(pars_dict,model_class, model_init_kwargs, x, return_model=False):
targets_names = model_init_kwargs['targets_names'] if 'targets_names' in model_init_kwargs.keys() else None
model_init_kwargs.update(pars_dict)
model = model_class(**model_init_kwargs)
if targets_names is not None:
if targets_names is not None and model_class not in [HoltWintersMulti, FK_multi]:
subscores = []
for c in targets_names:
model_init_kwargs.update({'target_name': c, 'targets_names': None})
Expand All @@ -32,13 +33,13 @@ def fit_sample(pars_dict,model_class, model_init_kwargs, x, targets_names=None,
subscores.append(score_autoregressive(model, x[features_names], target_name=c))
score = np.mean(subscores)
else:
score = score_autoregressive(model, x)
model.fit(x)
if return_model:
return model
else:
return score

def tune_hyperpars(x, model_class, hyperpars, n_trials=100, targets_names=None, verbose=True, return_model=True, parallel=True, **model_init_kwargs):
def tune_hyperpars(x, model_class, hyperpars, n_trials=100, verbose=True, return_model=True, parallel=True, **model_init_kwargs):
"""
:param x: pd.DataFrame (n, n_cov)
:param y: pd.Series (n)
Expand Down Expand Up @@ -71,14 +72,14 @@ def tune_hyperpars(x, model_class, hyperpars, n_trials=100, targets_names=None,
from multiprocessing import cpu_count
with ProcessPoolExecutor(max_workers=np.minimum(cpu_count(), 10)) as executor:
scores = list(tqdm(executor.map(partial(fit_sample, model_class=model_class,
model_init_kwargs=model_init_kwargs, x=x, targets_names=targets_names),
model_init_kwargs=model_init_kwargs, x=x),
pars_cartridge), total=n_trials, desc='Tuning hyperpars for {}'.format(model_class.__name__)))
if verbose:
plt.figure()
t = np.linspace(0.01, 0.99, 30)
plt.plot(np.quantile(scores, t), t)
else:
fitted_model = fit_sample(pars_cartridge[0], model_class, model_init_kwargs, x, targets_names=targets_names, return_model=True)
fitted_model = fit_sample(pars_cartridge[0], model_class, model_init_kwargs, x, return_model=True)

if model_class in [HoltWintersMulti]:
alphas = np.array([m.alpha for m in fitted_model.models])
Expand Down Expand Up @@ -384,7 +385,9 @@ def __init__(self, periods, target_name, targets_names=None, q_vect=None, val_ra
self.init_pars = {'periods': periods, 'target_name': target_name, 'q_vect': q_vect, 'val_ratio': val_ratio, 'nodes_at_step': nodes_at_step,
'optimization_budget': optimization_budget, 'n_sa': n_sa, 'constraints': constraints, 'optimize_hyperpars': optimize_hyperpars,
'verbose':verbose, 'alphas': alphas, 'gammas_1': gammas_1, 'gammas_2': gammas_2,
'models_periods': models_periods, 'optimize_submodels_hyperpars':optimize_submodels_hyperpars}
'models_periods': models_periods, 'optimize_submodels_hyperpars':optimize_submodels_hyperpars,
'targets_names': targets_names}

self.targets_names = [target_name] if targets_names is None else targets_names
self.periods = periods
self.optimization_budget = optimization_budget
Expand All @@ -411,7 +414,7 @@ def reinit_pars(self):

def fit(self, x_pd, y_pd=None):
if self.optimize_hyperpars:
pars_opt = tune_hyperpars(x_pd, HoltWintersMulti, hyperpars={'fake':[0, 1]}, n_trials=1, targets_names=self.targets_names, return_model=False, parallel=False, **self.init_pars)
pars_opt = tune_hyperpars(x_pd, HoltWintersMulti, hyperpars={'fake':[0, 1]}, n_trials=1, return_model=False, parallel=False, **self.init_pars)
self.set_params(**pars_opt)
self.reinit_pars()

Expand Down Expand Up @@ -796,10 +799,11 @@ def __init__(self, target_name='target', targets_names=None, n_sa=1, n_predicto
n_predictors = int(n_predictors)
self.base_predictor = base_predictor
self.targets_names = targets_names
self.init_pars = {'target_name': target_name, 'n_sa': n_sa, 'n_predictors': n_predictors, 'alpha': alphas, 'm': m, 'omega': omegas,
self.init_pars = {'target_name': target_name, 'n_sa': n_sa, 'n_predictors': n_predictors, 'alpha': alphas, 'm': m, 'omegas': omegas,
'ns_harmonics': ns_harmonics, 'val_ratio': val_ratio, 'nodes_at_step': nodes_at_step,
'q_vect': q_vect, 'periodicity': periodicity, 'optimize_hyperpars': optimize_hyperpars,
'optimization_budget': optimization_budget, 'r': r, 'q': q,'optimize_submodels_hyperpars':optimize_submodels_hyperpars}
'optimization_budget': optimization_budget, 'r': r, 'q': q,'optimize_submodels_hyperpars':optimize_submodels_hyperpars,
'verbose':verbose, 'base_predictor': base_predictor,'targets_names':targets_names}
self.init_pars.update(scengen_kwgs)
self.verbose=verbose
self.optimize_hyperpars = optimize_hyperpars
Expand All @@ -811,7 +815,7 @@ def __init__(self, target_name='target', targets_names=None, n_sa=1, n_predicto
if self.periodicity < n_sa:
print('WARNING: periodicity is smaller than n_sa, this may lead to suboptimal results.')
self.alphas = np.ones(n_predictors) * 0.01 if alphas is None else alphas
self.omegas = np.ones(n_predictors) * 0.01 if alphas is None else alphas
self.omegas = np.ones(n_predictors) * 0.01 if omegas is None else omegas
self.n_sa = n_sa
self.m = m
self.target_name = target_name
Expand Down Expand Up @@ -854,7 +858,7 @@ def reinit_pars(self):
def fit(self, x_pd, y_pd=None, **kwargs):
if self.optimize_hyperpars:
pars_opt = tune_hyperpars(x_pd, FK_multi, hyperpars={'fake':[0, 1]},
n_trials=1, targets_names=self.targets_names, return_model=False, parallel=False,
n_trials=1, return_model=False, parallel=False,
**self.init_pars)
self.set_params(**pars_opt)
self.reinit_pars()
Expand Down Expand Up @@ -885,7 +889,7 @@ def run(self, x_pd, return_coeffs=True, fit=True):
# set stored states
if fit:
for j in range(self.n_predictors):
self.models[j].fit(x_pd, x_pd[self.target_name])
self.models[j].fit(x_pd)

preds_experts = np.dstack([ self.models[j].predict(x_pd) for j in range(self.n_predictors)])

Expand Down
6 changes: 3 additions & 3 deletions tests/test_models.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -100,11 +100,11 @@ def test_hw_multi(self):
self.data = self.data.resample('1h').mean()
df_tr, df_te = self.data.iloc[:1200], self.data.iloc[1200:1500]
steps_day = 24
fks_multi = FK_multi(n_predictors=6, n_sa=steps_day, m=steps_day*7,
fks_multi = FK_multi(n_predictors=2, n_sa=steps_day, m=steps_day*7,
target_name='all', periodicity=steps_day*2,
optimize_hyperpars=True, optimization_budget=2, targets_names=['all']).fit(df_tr, df_tr['all'])
optimize_hyperpars=True, optimization_budget=2, targets_names=df_tr.columns[:2]).fit(df_tr)
hw_multi = HoltWintersMulti(periods=[steps_day, steps_day * 7], n_sa=steps_day, optimization_budget=100, q_vect=np.arange(11) / 10,
target_name='all', models_periods=np.array([1,2,steps_day])).fit(df_tr, df_tr['all'])
target_name='all', models_periods=np.array([1,2,steps_day]), targets_names=df_tr.columns[:6]).fit(df_tr)


fks = FK(n_sa=steps_day, m=steps_day, target_name='all',
Expand Down

0 comments on commit 60b597f

Please sign in to comment.