util.py

# encoding: utf-8
import numpy as np
import pandas as pd

def cml(fns, xs, ys):
    """ Cuadrados minimos lineales.

        * fns: una función de R -> R^n donde el vector
        de salida son las distintas funciones lineales.

        * xs, ys: los vectores de datos

        Devuelve una función R -> R
    """
    A = np.array([fns(x) for x in xs], dtype=np.float64)
    ws = np.linalg.lstsq(A, ys)[0]
    return lambda x: sum([w*y for w, y in zip(ws, fns(x))])

def poly(grado):
    """ Genera un polinomio grado `grado` preparado
        para pasarle a `cml`.
    """
    return lambda x: [float(x)**i for i in range(grado, -1, -1)]

def ecm(xs, ys):
    """ Error cuadratico medio.
        Toma xs e ys vectores de predicho y real respectivamente.
    """
    if len(xs) != len(ys): raise Exception('Different size!')
    return (sum([(x - y)**2 for x, y in zip(xs, ys)]))/len(xs)

def colapsarK(df, column, k, drop=True):
    for i in range(1, k+1):
        df['prev' + str(i)] = df[column].shift(i)

    if drop:
        toDrop = list(range(0, k))
        return df.drop(df.index[toDrop]).reset_index(drop=True)
    else:
        return df

def ar(cant):
    """ Genera la funcion de AR
    """
    if cant < 1: raise Exception('Value must be greater than 0')
    return lambda x: [1] + [x[i] for i in range(cant+1)]