How to rebalance portfolio with complext logic, I found size only allow scalar .

[eromoe]

Hi, I am testing vectorbt, I would like purchase after comfirming it fit all my need.
I have scrolled the whole issue list, only Portfolio.from_order_func can do complex logic.

My test logic is

1. every 100days select 16 stocks
2. every 10days use RSI to rebalance selected stocks
3. combine above two signal, no row contain more than 9 stocks
4. set each stock value as 1/10 of total assets. ( if there is no True stock, just keep an empty position)

I have generate singlas in step 1,2,3 , but 4 got trouble

My test code is as below

import pandas as pd
import numpy as np
from numba import njit

import vectorbt as vbt
from vectorbt.portfolio import nb
from vectorbt.portfolio.enums import Direction, SizeType


np.random.seed(42)

close = pd.DataFrame(data=np.repeat([1, 2, 3, 4, 5],3).reshape(5,3), columns=list('abc'))
# sumilate 1,2,3 step output,  in real shape is (1100, 2500)  and mark all valid date/stock True.
entries = np.random.randint(0,2, size=(5,3))

@njit
def order_func_nb(c, size_type, direction):
    size = np.zeros_like(entries[0], type=np.float_)

    for col in range(c.from_col, c.to_col):
        size[col] = entries[c.i, col] * 0.1
        
    return nb.order_nb(size=size, size_type=size_type, direction=direction)


pf = vbt.Portfolio.from_order_func(
    close,
    order_func_nb,
    vbt.Rep('size_type'),
    vbt.Rep('direction'),
    broadcast_named_args=dict(
        entries=entries,
        size_type=SizeType.TargetPercent,
        direction=Direction.LongOnly,
    ),
)

It complain that

TypeError: only size-1 arrays can be converted to Python scalars

nb.order_nb only accept scalar . So how do I achieve my goals ?

[eromoe]

After reread docs, I found

The frame is further divided into "blocks": columns, groups, rows, segments, and elements.
columns can be grouped into groups that may or may not share the same capital. Regardless of capital sharing, each collection of elements within a group anda time step is called a segment, which simply defines a single context (such as shared capital) for one or multiple orders. Each segment can also define a custom sequence (a so-called call sequence) in which orders are executed.

The way I do:

1. use segment_mask to read size every 10 tick
2. in pre_segment_func_nb , check wether c.i//100==0 in order to change stock every 100 tick.

Seems correct, don't know is it recommended .