While writing my own ta library, I noticed a discrepancy between my OBV values and ta's.

I did some digging and noticed 2 easy to fix issues. I've ran the unit tests and everything still works 👍

Issues:

1. For the initial volume value, we apply 0 because there is no history to confirm whether the movement was up or down.
2. When the current days close is equal to the previous, we apply 0 because there was no up or down movement.

Fixes:

1. Initialize an empty zeros array with the same shape as the volume with np.zeros_like()
2. Use masks to insert positive and negative volume values into the zeros array.

Reasoning:

1. We can see in the screenshot below, that the initial day has no obv value. This is effectively 0
   
2. In the Investopedia formula, we can see if close = close_prev then 0, this is not handled properly in the current implementation.
   

Tests: