[wavebyrd]

FractionalBacktest.run() scales EntryPrice, ExitPrice, SL, TP and the overlay indicators back up after the run, but the Entry_<indicator> / Exit_<indicator> columns in stats._trades stayed at the fractional scale, so they no longer matched the prices next to them:

from backtesting.lib import FractionalBacktest
from backtesting.test import BTCUSD
from backtesting.test._test import SmaCross

stats = FractionalBacktest(BTCUSD['2015':], SmaCross, fractional_unit=1/1e6, cash=100,
                           finalize_trades=True).run(fast=2, slow=3)
t = stats['_trades'].iloc[0]
print(t['EntryPrice'], t['Entry_SMA(C,2)'])  # 236.69 0.00023414, now 236.69 234.14

Those columns are now scaled for overlay indicators too, using the same column names as _stats.py (including the _0, _1 suffixes for multi-dimensional indicators). Added an assertion to test_FractionalBacktest, which fails without the fix; the rest of the suite and flake8 pass.

[arhancanli]

I ran this branch (pull/1426/head, Python 3.12, current pandas) against a strategy with a one-dimensional overlay (SMA), a two-row overlay (name='band', rolling max over rolling min) and a non-overlay indicator (name='ret', a percent change), with fractional_unit=1/1e6 on BTCUSD['2015':].

First trade, entry price 261.7:

Entry_SMA(C,2)  273.945     Entry_SMA(C,3)  259.803
Entry_band_0    286.11      Entry_band_1    261.78
Entry_ret       9.294

The overlay columns now sit in the same units as EntryPrice, including the _0/_1 suffixed ones, which matches the suffix rule in _stats.py lines 82-85 (len(ind) > 1). The non-overlay ret column is untouched, which is what you want: it is a percentage, not a price.

The trades.columns.intersection(cols) guard also covers Entry_/Exit_ columns that are absent when there are no trades (if len(trades_df) and strategy_instance in _stats.py), so an empty run does not raise.

One thing I did not test and the diff doesn't settle: a non-overlay indicator that is still price-denominated, such as an ATR or a High - Low spread. Before this PR those were already left in fractional units in _indicators, and overlay is the only switch the class looks at, so the new trade columns stay consistent with the plotted arrays. If that is intended, a sentence in the FractionalBacktest docstring ("only overlay indicators are rescaled") would save the next reader from finding it by surprise.

[arhancanli]

Ran this on f403424 (Python 3.12, BTCUSD from 2015, fractional_unit=1/1e6) with a strategy using two overlay SMAs, an SL, and one non-overlay indicator (overlay=False, constant 1). On master, Entry_SMA(C,2) for trade 0 is 0.000274 next to EntryPrice 261.7. With this branch it is 273.945, Exit_SMA(C,3) comes out as 259.82 and the non-overlay column stays 1.0, so the overlay filter is respected.

One thing I could not tell from the diff: the multi-dimensional case (_0, _1 suffixes). The test only covers the single-output SMA. Is there an existing indicator that returns a 2-row array with overlay=True that could be added to the test, so the suffix branch is exercised too?