import pandas as pd
import numpy as np

# convert 1-dimensional hourly time series into numpy array with 24 columns and corresponding number of days (rows)
# option to repeat first and last hours N times everyday (used in clustering)
def col_to_M(v_column, repeat=None):
    
    if isinstance(v_column, pd.Series):
        N_days=int(v_column.count()/24) # to take leap year into account
        series=v_column.values
    elif isinstance(v_column, np.ndarray):
        N_days=int(len(v_column)/24)
        series=v_column
    else:
        raise ValueError("Input vector must be a pandas series or numpy array.")
    
    if repeat is None:
        return series.reshape(N_days, 24)

    elif isinstance(repeat,int):
        # repeat first and last hours to increase their weights
        M=series.reshape(N_days, 24)
        return np.concatenate([np.repeat(M[:,0].reshape(N_days,1), repeat, axis=1), M, np.repeat(M[:,-1].reshape(N_days,1), repeat, axis=1)], axis=1)
    
    else:
        raise ValueError("The number of repeats must be an integer!")

# select valid draws by start and end values (for continuity)
def get_valid_draws(M_original , M_draw, epsilon=0.05):
    
    if M_draw.ndim == 1:
        M_valid = M_draw
    
    elif M_draw.ndim == 2 and M_draw.shape[0]<=5:
        M_valid = M_draw
        
    else: 
        start_mean=M_original[:,0].mean()
        end_mean=M_original[:,-1].mean()

        M_valid=M_draw[np.where((M_draw[:,0]<start_mean+epsilon) & (M_draw[:,0]>start_mean-epsilon) &
                    (M_draw[:,-1]<end_mean+epsilon) & (M_draw[:,-1]>end_mean-epsilon))]
        
        # valid draws must also have all values between 0 and 1 (1.1 to allow for scaling back down to original maximum later on) 
        M_valid=M_valid[np.all((M_valid>=0) & (M_valid<=1.1), axis=1)]
    
    if M_valid.size==0:
        M_valid=M_original.mean(axis=0)
    
    return M_valid