import statsmodels.api as sm
import numpy as np

def fit_by_lowess(x, y):
    data=np.column_stack((x,y))
    data = data[data[:, 0] != 0]
    x,y=data[:,0],data[:,1]

    # 挑选合适frac
    frac=choose_frac(x,y)

    # 拟合 by lowess
    lowess = sm.nonparametric.lowess
    y_lowess = lowess(y, x, frac)
    x_fit, y_fit = zip(*y_lowess)
    x_fit, y_fit = np.array(x_fit), np.array(y_fit)
    return x_fit, y_fit


def choose_frac(x,y,delta_k_threshold=3):
    import statsmodels.api as sm
    class InputDataError(Exception):
        def __init__(self, message) -> None:
            super().__init__(message)

    for frac in np.linspace(0,1,20)[2:]:
        lowess = sm.nonparametric.lowess(y, x, frac)  
        x_smoothed,y_smoothed =map(np.array,zip(*lowess)) 
        
        
        delta_x=x_smoothed[1:]-x_smoothed[:-1]
        delta_x = np.where(delta_x == 0, np.inf,delta_x)

        delta_y=y_smoothed[1:]-y_smoothed[:-1]
        
        k_list=delta_y/delta_x

        k_delta=k_list[1:]-k_list[:-1]
        if abs(np.max(k_delta)) < delta_k_threshold:
            return frac
    else:
        raise InputDataError("Please check if your filtered data is approiate")

class Predictor:
    def __init__(self,x_fit,y_fit) -> None:
        def make_x_unique(x_fit,y_fit):
            data=np.column_stack((x_fit,y_fit))
            unique_data=dict()
            for x,y in data:
                if not unique_data.get(x):
                    unique_data[x]=y
                else:
                    pass
            return np.array(list(unique_data.items()))
        
        from scipy.interpolate import interp1d
        data=make_x_unique(x_fit,y_fit)
        x_fit,y_fit=data[:,0],data[:,1]
        self.f = interp1d(x_fit, y_fit, kind='linear', fill_value='extrapolate')

    def predict(self,x_interp):
         return self.f(x_interp)