# Copyright (c) 2021. Bram Vonk, Enexis import numpy as np import pandas as pd [docs]def downcast(s, try_numeric=True, category=False): """ Downcast a series to the lowest possible memory type. Parameters ---------- s : pd.Series Series to downcast. try_numeric: bool If True it will try to read strings as numeric values. category: bool If True (string) objects will be cast as a category. Returns ------- Downcasted series. """ if try_numeric: if s.dtype.kind != "M": s = pd.to_numeric(s, errors="ignore") if category: if s.dtype.kind == "O": s = s.astype("category") if s.dtype.kind == "f": s = pd.to_numeric(s, errors="ignore", downcast="float") elif s.dtype.kind == "i": s = pd.to_numeric(s, errors="ignore", downcast="signed") s = pd.to_numeric(s, errors="ignore", downcast="unsigned") return s [docs]def map_labels(series, kind="categorical", labels=None, backwards=False, **arg): """ Map a Series values by the labels given. Parameters ---------- series: pd.Series Series to map on. kind: str Indicator for kind of preprocess in series. With kind of {"categorical", "ordinal"} the mapping is applied, otherwise not. labels: dict Defines with the mapping {key_0: value_0, etc.}. arg: Additional arguments. Returns ------- pd.Series Series with mapped values. """ if kind in ["categorical", "ordinal"]: if isinstance(labels, dict): if backwards: labels = {v: k for k, v in labels.items()} series = series.map(labels) return series [docs]class MinMaxScaler: """MinMax Scaler like in sklearn, prevents total library import/dependency.""" def __init__(self, upper=1, lower=-1): """ Initialize scaler with upper and lower boundary. Parameters ---------- upper : float upper boundary to scale to lower : float lower boundary to scale to """ self.upper = upper self.lower = lower self.max = None self.min = None [docs] def fit(self, X, y=None): """ Get fit parameters. Parameters ---------- X : np.array preprocess to fit on y : None solely for consistency Returns ------- self instance with self.min, self.max defined. """ self.max = np.max(X) self.min = np.min(X) return self [docs] def transform(self, X): """ Scales preprocess according to fitted parameters. Parameters ---------- X : np.array preprocess to scale Returns ------- np.array scaled preprocess """ X_01 = (X - self.min) / (self.max - self.min) return X_01 * (self.upper - self.lower) + self.lower [docs] def fit_transform(self, X, y=None): """ Execute consecutively self.fit and self.transform. Parameters ---------- X : np.array preprocess to scale y : None solely for consistency Returns ------- np.array scaled preprocess """ return self.fit(X).transform(X) [docs] def inverse_transform(self, X, y=None): """ Scale back to original domain. Parameters ---------- X : np.array preprocess to scale y : None solely for consistency Returns ------- np.array scaled preprocess """ X_xx = (X - self.lower) / (self.upper - self.lower) return X_xx * (self.max - self.min) + self.min