Support iterative take profit or loss signals about vectorbt

[polakowo]

@NowBothWhiteAndRed
Here is an example how to do just that:

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

@njit
def exit_choice_func_nb(col, from_i, to_i, ts, sl_stops, tp_stops, trailing):
    # Get indices of stop loss and take profit exits and return the first
    sl_idxs = vbt.signals.nb.sl_choice_nb(col, from_i, to_i, ts, sl_stops, trailing, True)
    tp_idxs = vbt.signals.nb.tp_choice_nb(col, from_i, to_i, ts, tp_stops, True)
    return np.sort(np.concatenate((sl_idxs, tp_idxs)))[:1]

@njit
def custom_iter_apply_nb(i, entries, ts, sl_stops, tp_stops, trailing):
    # Run exit_choice_func_nb for stop at index i
    return vbt.signals.nb.generate_enex_nb(
        entries.shape,
        vbt.signals.nb.true_choice_nb,
        exit_choice_func_nb,
        (entries,), (ts, sl_stops[i, :, :], tp_stops[i, :, :], trailing))

@njit
def generate_custom_exits_nb(entries, ts, sl_stops, tp_stops, trailing=False):
    # Run custom_iter_apply_nb for each stop value and concatenate results
    return vbt.base.combine_fns.apply_and_concat_multiple_nb(
        len(sl_stops), custom_iter_apply_nb, entries, ts, sl_stops, tp_stops, trailing)

def generate_custom_exits(entries, price, sl_stops, tp_stops, trailing=False, 
                          keys=None, broadcast_kwargs={}):
    vbt.utils.checks.assert_type(price, (pd.Series, pd.DataFrame))
    # Broadcast pandas objects
    entries, price = vbt.base.reshape_fns.broadcast(entries, price, **broadcast_kwargs, writeable=True)
    # Broadcast stop values for each to match the shape of entries
    sl_stops = vbt.base.reshape_fns.broadcast_to_array_of(sl_stops, entries.vbt.to_2d_array())
    tp_stops = vbt.base.reshape_fns.broadcast_to_array_of(tp_stops, entries.vbt.to_2d_array())
    # Broadcast stop values between each other
    sl_stops, tp_stops = np.broadcast_arrays(sl_stops, tp_stops)
    sl_stops = np.copy(sl_stops)
    tp_stops = np.copy(tp_stops)
    # Build column hierarchy
    if keys is not None:
        param_columns = keys
    else:
        sl_param_columns = vbt.base.index_fns.index_from_values(sl_stops, name='stop_loss')
        tp_param_columns = vbt.base.index_fns.index_from_values(tp_stops, name='take_profit')
        param_columns = vbt.base.index_fns.stack_indexes(sl_param_columns, tp_param_columns)
    columns = vbt.base.index_fns.combine_indexes(param_columns, entries.vbt.columns)
    # Execute
    new_entries, exits = generate_custom_exits_nb(
        entries.vbt.to_2d_array(),
        price.vbt.to_2d_array(),
        sl_stops, 
        tp_stops, 
        trailing
    )
    # Wrap numpy arrays into pandas objects and return
    return entries.vbt.wrap(new_entries, columns=columns), entries.vbt.wrap(exits, columns=columns)

entries = pd.DataFrame({
    'a': [True, False, False, False, False],
    'b': [True, False, True, False, True],
    'c': [True, True, True, False, False]
})
price = pd.Series([1., 2., 3., 2., 1.])
generate_custom_exits(entries, price, 0.1, 0.15)

# stop_loss      0.1              
# take_profit    0.1              
#                  a      b      c
# 0             True   True   True
# 1            False  False  False
# 2            False   True   True
# 3            False  False  False
# 4            False   True  False

# stop_loss      0.1              
# take_profit    0.1              
#                  a      b      c
# 0            False  False  False
# 1             True   True   True
# 2            False  False  False
# 3            False   True   True
# 4            False  False  False

While the code seems complex, the only thing that I created was the exit_choice_func_nb that combines signals using OR. All other functions are what vectorbt is doing under the hood: broadcasting inputs and parameters, preparing inputs for numba, iterating over parameters and concatenating the results, and wrapping the results back into pandas objects. So the generate_custom_exits function is the most complete you can get.

[polakowo]

Maybe there is a need to integrate such use cases into the library but for now I'm more concerned with having multi-asset portfolios in vectorbt.

@TheSnowGuru it won't work since stop values are not absolute values but percentages relative to the price at the entry point. There is no (at least efficient) way of deciding which came first without utilizing numba.

[NowBothWhiteAndRed]

Thanks! The example code works as expected. It helped me a lot.

This case could be generalized to "chain any signal with iterative logic with others" so it would definitely be worthwhile to support for building more advanced strategies.

[polakowo]

Implemented in #52