def generate_standard_deviation_data(
self, model: RegressorMixin) -> RegressorMixin:
# Define some variables
predicted = None
self.standard_deviation = pd.DataFrame()
# Loop over horizons and ask prediction for each specific horizon
for horizon in self.validation_data.horizon.unique():
# Make subset for this specific horizon
sub_val = self.validation_data[self.validation_data.horizon ==
horizon]
try:
predicted = model.predict(sub_val.iloc[:, 1:-1])
except Exception as e:
print("Could not get prediction from new model!", e)
# Calculate confidence interval for this horizon
confidence_interval_horizon = self._calculate_standard_deviation(
sub_val.iloc[:, 0], predicted)
confidence_interval_horizon[
"horizon"] = horizon # Label with respective horizon
self.standard_deviation = pd.concat(
[self.standard_deviation, confidence_interval_horizon])
model.standard_deviation = self.standard_deviation
return model
def __init__(self, binner=None, estimator=None, n_jobs=None, verbose=False):
"""
@param binner transformer or predictor which creates the buckets
@param estimator predictor trained on every bucket
@param n_jobs number of parallel jobs (for training and predicting)
@param verbose boolean or use ``'tqdm'`` to use :epkg:`tqdm`
to fit the estimators
*binner* allows the following values:
- ``tree``: the model is :epkg:`sklearn:tree:DecisionTreeRegressor`
- ``'bins'``: the model :epkg:`sklearn:preprocessing:KBinsDiscretizer`
- any instanciated model
*estimator* allows the following values:
- ``None``: the model is :epkg:`sklearn:linear_model:LinearRegression`
- any instanciated model
"""
if estimator is None:
estimator = LinearRegression()
if binner in ('tree', None):
binner = DecisionTreeRegressor(min_samples_leaf=2)
RegressorMixin.__init__(self)
PiecewiseEstimator.__init__(self, binner=binner, estimator=estimator,
n_jobs=n_jobs, verbose=verbose)
def __init__(self, num_inputs, mxseed=0, epochs=5000, net_type=1):
BaseEstimator.__init__(self)
RegressorMixin.__init__(self)
self.net = None
self.num_inputs = num_inputs
self.mxseed = mxseed
self.epochs = epochs
self.net_type = net_type
return
def valscore(self, Xn, yn, scoring):
if scoring == 'weighted':
return (RegressorMixin.score(self,
Xn,
yn,
sample_weight=self.caseweights_))
elif scoring == 'normal':
return (RegressorMixin.score(self, Xn, yn))
else:
ValueError('Scoring flag must be set to "weighted" or "normal".')
def valscore(self,Xn,yn,scoring):
n,p,Xn = _predict_check_input(Xn)
(n,p) = Xn.shape
if p!= self.X.shape[1]:
raise(ValueError('New data must have seame number of columns as the ones the model has been trained with'))
if scoring=='weighted':
return(RegressorMixin.score(self,Xn,yn,sample_weight=self.caseweights_))
elif scoring=='normal':
return(RegressorMixin.score(self,Xn,yn))
else:
raise(ValueError('Scoring flag must be set to "weighted" or "normal".'))
def __init__(self, model='SIR', t=0, max_iter=100,
learning_rate_init=0.1, lr_schedule='constant',
momentum=0.9, power_t=0.5, early_th=None,
min_threshold='auto', max_threshold='auto',
verbose=False, init=None):
if init is not None:
if isinstance(init, EpidemicRegressor):
if hasattr(init, 'coef_'):
init = init.coef_.copy()
else:
init = None # pragma: no cover
elif not isinstance(init, dict):
raise TypeError(
f"init must be a dictionary not {type(init)}.")
BaseEstimator.__init__(self)
RegressorMixin.__init__(self)
self.t = t
self.model = model
self.max_iter = max_iter
self.learning_rate_init = learning_rate_init
self.lr_schedule = lr_schedule
self.momentum = momentum
self.power_t = power_t
self.early_th = early_th
self.verbose = verbose
if min_threshold == 'auto':
if model.upper() in ('SIR', 'SIRD'):
min_threshold = 0.0001
elif model.upper() in ('SIRC', ):
pmin = dict(beta=0.001, nu=0.0001, mu=0.0001,
a=-1., b=0., c=0.)
min_threshold = numpy.array(
[pmin[k[0]] for k in CovidSIRDc.P0])
elif model.upper() in ('SIRDC'):
pmin = dict(beta=0.001, nu=0.001, mu=0.001,
a=-1., b=0., c=0.)
min_threshold = numpy.array(
[pmin[k[0]] for k in CovidSIRDc.P0])
if max_threshold == 'auto':
if model.upper() in ('SIR', 'SIRD'):
max_threshold = 1.
elif model.upper() in ('SIRC', 'SIRDC'):
pmax = dict(beta=1., nu=0.5, mu=0.5,
a=0., b=4., c=2.)
max_threshold = numpy.array(
[pmax[k[0]] for k in CovidSIRDc.P0])
self.min_threshold = min_threshold
self.max_threshold = max_threshold
self._get_model()
self.init = init
if init is not None:
self.coef_ = init
def __init__(self, force_positive=False, **kwargs):
"""
*kwargs* should contains parameters
for :epkg:`sklearn:decomposition:NMF`.
The parameter *force_positive* removes all
negative predictions and replaces by zero.
"""
BaseEstimator.__init__(self)
RegressorMixin.__init__(self)
MultiOutputMixin.__init__(self)
for k, v in kwargs.items():
setattr(self, k, v)
self.force_positive = force_positive
def valscore(self,Xn,yn,scoring):
if type(Xn) == ps.core.frame.DataFrame:
Xn = Xn.to_numpy()
if type(yn) in [ps.core.frame.DataFrame,ps.core.series.Series]:
yn = yn.to_numpy().T.astype('float64')
(n,p) = Xn.shape
if p!= self.X.shape[1]:
raise(ValueError('New data must have seame number of columns as the ones the model has been trained with'))
if scoring=='weighted':
return(RegressorMixin.score(self,Xn,yn,sample_weight=self.caseweights_))
elif scoring=='normal':
return(RegressorMixin.score(self,Xn,yn))
else:
raise(ValueError('Scoring flag must be set to "weighted" or "normal".'))
def _fit_and_predict_oof_model(
self,
estimator: RegressorMixin,
X: ArrayLike,
y: ArrayLike,
train_index: ArrayLike,
val_index: ArrayLike,
sample_weight: Optional[ArrayLike] = None,
) -> Tuple[RegressorMixin, NDArray, ArrayLike]:
"""
Fit a single out-of-fold model on a given training set and
perform predictions on a test set.
Parameters
----------
estimator : RegressorMixin
Estimator to train.
X : ArrayLike of shape (n_samples, n_features)
Input data.
y : ArrayLike of shape (n_samples,)
Input labels.
train_index : ArrayLike of shape (n_samples_train)
Training data indices.
val_index : ArrayLike of shape (n_samples_val)
Validation data indices.
sample_weight : Optional[ArrayLike] of shape (n_samples,)
Sample weights. If None, then samples are equally weighted.
By default ``None``.
Returns
-------
Tuple[RegressorMixin, NDArray, ArrayLike]
- [0]: RegressorMixin, fitted estimator
- [1]: NDArray of shape (n_samples_val,),
estimator predictions on the validation fold.
- [3]: ArrayLike of shape (n_samples_val,),
validation data indices.
"""
X_train = _safe_indexing(X, train_index)
y_train = _safe_indexing(y, train_index)
X_val = _safe_indexing(X, val_index)
if sample_weight is None:
estimator = fit_estimator(estimator, X_train, y_train)
else:
sample_weight_train = _safe_indexing(sample_weight, train_index)
estimator = fit_estimator(
estimator, X_train, y_train, sample_weight_train
)
if _num_samples(X_val) > 0:
y_pred = estimator.predict(X_val)
else:
y_pred = np.array([])
return estimator, y_pred, val_index
def __init__(self, rf_estimator=None, lasso_estimator=None):
"""
@param rf_estimator random forest estimator,
:epkg:`sklearn:ensemble:RandomForestRegressor`
by default
@param lass_estimator Lasso estimator,
:epkg:`sklearn:linear_model:LassoRegression`
by default
"""
BaseEstimator.__init__(self)
RegressorMixin.__init__(self)
if rf_estimator is None:
rf_estimator = RandomForestRegressor()
if lasso_estimator is None:
lasso_estimator = Lasso()
self.rf_estimator = rf_estimator
self.lasso_estimator = lasso_estimator
def produce_submission(model: RegressorMixin):
td = load_test_data()
out: pd.DataFrame = model.predict(td)
submission_data = pd.DataFrame([td.index, out]).T
submission_data.columns = ['Id', 'SalePrice']
submission_data = submission_data.astype({'Id': int, 'SalePrice': float})
submission_data['SalePrice'].round(decimals=2)
submission_data.to_csv('../data/submissions/nearest-neighbors.csv', header=['Id', 'SalePrice'], index=False)
def _train(train_data: DataFrame,
regressor: RegressorMixin,
clusterer: Clustering,
do_cv=False) -> dict:
models = dict()
train_data = clusterer.cluster_data(train_data)
for cluster in range(clusterer.n_clusters):
cluster_train_df = train_data[cluster]
if not cluster_train_df.empty:
cluster_targets_df = cluster_train_df['label']
if do_cv:
cross_validation_result = cross_validate(
regressor,
cluster_train_df.drop('label', 1),
cluster_targets_df.values.ravel(),
return_estimator=True,
cv=10 #TODO per Chiara check se vuoi 10 cv
)
validation_scores = cross_validation_result['test_score']
regressors = cross_validation_result['estimator']
regressor = regressors[dict(
zip(validation_scores, range(len(validation_scores)))
)[max(
validation_scores
)]] #TODO per Chiara check se vuoi il max o min o quello che sta in mezzo
else:
regressor.fit(cluster_train_df.drop('label', 1),
cluster_targets_df.values.ravel())
models[cluster] = regressor
try:
regressor = clone(regressor)
except TypeError:
regressor = clone(regressor, safe=False)
return {
ModelType.CLUSTERER.value: clusterer,
ModelType.REGRESSOR.value: models
}
def __init__(self, estimator=None, n_estimators=10, n_jobs=None,
alpha=1., verbose=False):
"""
@param estimator predictor trained on every bucket
@param n_estimators number of estimators to train
@param n_jobs number of parallel jobs (for training and predicting)
@param alpha proportion of samples resampled for each training
@param verbose boolean or use ``'tqdm'`` to use :epkg:`tqdm`
to fit the estimators
"""
BaseEstimator.__init__(self)
RegressorMixin.__init__(self)
if estimator is None:
raise ValueError("estimator cannot be null.")
self.estimator = estimator
self.n_jobs = n_jobs
self.alpha = alpha
self.verbose = verbose
self.n_estimators = n_estimators
def _train(train_data: DataFrame, regressor: RegressorMixin, clusterer: Clustering) -> dict:
models = dict()
train_data = clusterer.cluster_data(train_data)
for cluster in range(clusterer.n_clusters):
cluster_train_df = train_data[cluster]
if not cluster_train_df.empty:
cluster_targets_df = cluster_train_df['label']
regressor.fit(cluster_train_df.drop('label', 1), cluster_targets_df.values.ravel())
models[cluster] = regressor
try:
regressor = clone(regressor)
except TypeError:
regressor = clone(regressor, safe=False)
return {'clusterer': clusterer, PredictiveModels.REGRESSION.value: models}
def _cv_estimate(model: RegressorMixin,
train_data: pd.DataFrame,
features: List[str],
y: str,
n_splits: int) -> Tuple[pd.Series, List[RegressorMixin]]:
cv = KFold(n_splits=n_splits)
models = []
cv_pred = pd.Series(np.nan, index=train_data.index)
for train, test in cv.split(train_data):
m = model.fit(train_data[features].iloc[train], train_data[y].iloc[train])
cv_pred.iloc[test] = m.predict(train_data[features].iloc[test])
models += [m]
return cv_pred, models
def bootstrap_mae(
f: RegressorMixin,
X, # numpy array
y, # numpy array
num_samples: int = 100,
random_state: int = random.randint(0, 2 ** 32 - 1),
) -> List[float]:
"""
Take the regressor ``f``, and compute it's bootstrapped mse over the dataset ``X``,``y``.
Generate ``num_samples`` samples; and seed the resampler with ``random_state``.
"""
dist: List[float] = []
y_pred = f.predict(X) # type:ignore (predict not on ClassifierMixin)
# do the bootstrap:
for trial in range(num_samples):
sample_pred, sample_truth = resample(
y_pred, y, random_state=trial + random_state
) # type:ignore
score = mean_absolute_error(y_true=sample_truth, y_pred=sample_pred) # type:ignore
dist.append(score)
return dist
def bootstrap_regressor(
f: RegressorMixin,
X, # numpy array
y, # numpy array
num_samples: int = 100,
random_state: int = random.randint(0, 2 ** 32 - 1),
) -> List[float]:
"""
Take the regressor f, and compute it's bootstrapped accuracy over the dataset `X`,`y`.
Generate `num_samples` samples; and seed the resampler with `random_state`.
"""
dist: List[float] = []
y_pred = f.predict(X) # type:ignore
# do the bootstrap:
for trial in range(num_samples):
sample_pred, sample_truth = resample(
y_pred, y, random_state=trial + random_state
) # type:ignore
score = mean_squared_error(y_true=sample_truth, y_pred=sample_pred) # type:ignore
dist.append(score)
return dist
def __init__(self):
RegressorMixin.__init__(self)
BaseEstimator.__init__(self)
def _reg():
return RegressorMixin()
def score(self, X, y, sample_weight=None):
if len(X) == 1:
output = self.predict(X)
return output.shape[0] * mean_squared_error(y, output)
return RegressorMixin.score(self, X, y, sample_weight=sample_weight)
def score(
self, X: np.ndarray, y: np.ndarray, sample_weight: Optional[np.ndarray] = None
) -> float:
return RegressorMixin.score(self, X, y, sample_weight)
def score(self, X, y, sample_weight=None):
return RegressorMixin.score(self, X, y, sample_weight=sample_weight)
def __init__(self, base_estimator):
RegressorMixin.__init__(self)
BaseEstimator.__init__(self)
self.base_estimator = base_estimator