Rust MinMaxDownsampler seems to miss the global maxima

[LeiRui]

Hi, I run into the problem that the MinMaxDownsampler seems to miss the global maxima with the following test code:

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from tsdownsample import M4Downsampler,EveryNthDownsampler,LTTBDownsampler,MinMaxDownsampler # MinMaxLTTBDownsampler

def read_ucr(filename,col=0): # col starting from 0
    df=pd.read_csv(filename, header=None, delimiter=r"\s+")
    df2=df.T
    v=df2.iloc[:,col]
    v=v.to_numpy()
    t = np.linspace(0, 10, num=len(v))
    return t,v

filename="sample.txt"
t,v=read_ucr(filename,1)
print(max(v)) # 78.546175

s_ds = MinMaxDownsampler().downsample(t, v, n_out=4)
# Select downsampled data
t_ds = t[s_ds]
v_ds = v[s_ds]
print(max(v_ds)) # 27.700287

The original maximal value is 78.546175 while the downsampled maximal value is 27.700287. I don't know where the bug is. Hope to get some suggestions.

sample.txt

[jvdd]

Hi @LeiRui,

Thank you for bringing up this issue!

Upon investigation, it appears that the bug is caused by the numpy array not being contiguous. You can determine if a numpy array is contiguous by using the following code: v.flags['C_CONTIGUOUS']. This issue will not manifest itself if you create a new numpy array that contains the data of v (np.array(v)).

To fix this, I guess we should check if the array is contiguous - and when this is not the case, the scalar implementation should be executed.

We should also definitely include some tests for this behavior!

Cheers, Jeroen