For example, I have 1 minute scale tick data, and I'd like add 5 minute/ 15 minute feature upon .

I wrote a function

def apply_over(func, arr, stride):
	n = len(arr)
	s = np.empty(n).reshape(-1, stride)

	for i in range(stride):
		s[:,i] = func(arr[i::stride])

	return s.reshape(n,)

This generate 5 minute SMA features for each 1 minute tick .

arr = np.arange(n).astype(float)
sma_5_5 = apply_over(lambda a: talib.SMA(a, 5) , arr, 5)

you can think arr = np.arange(1000).astype(float) as a stock close price at minute level .
It is a 1000 time tick collection.

On every tick , I need calculate sma5 on

• 1 minute scale (simply talib.SMA(arr, 5) )
• 5 minute scale (apply_over(lambda a: talib.SMA(a, 5) , arr, 5) )
• 15 minute scale (apply_over(lambda a: talib.SMA(a, 5) , arr, 15) )

For example:
A series ...,500, 501, 502, 503, 504 , 505,....

At 505,

• 1 minute scale = SMA([ 501, 502, 503, 504 , 505])
• 5 minute scale = SMA([ 485, 490, 495, 500 , 505])

Using pandas resmple(freq='5min').first() would make gaps , shrink arr length from 1000 to 200 . You need do it 5 times with each shift [0,1,2,3,4] and apply SMA to make sure every tick have 5 minute feature, , like what I do in apply_over

Because I really like vectorbt, so share this idea here.