| Pages:
1
..
53
54
55
56 |
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
If anyone wants to try a pre-release, please let me know
GSB What's New (User Summary)
Covering roughly the last 30 days (v73.85 - v74.72)
=====================================================
1. ACCURACY: MANY FILTERS AND INDICATORS NOW MATCH EXACTLY
------------------------------------------------------------
A large effort went into making sure GSB's filters and indicators produce
the exact same results as the reference trading platforms they are
compared against (TradeStation and MultiCharts).
- A new "Match-Check" button was added so you can run a one-click
comparison of GSB's results against the reference platforms directly
from the Tester Settings screen.
- The match-check report was improved several times: it now shows the
worst mismatches first, includes notes explaining likely causes,
summarizes price data and settings used, and ignores meaningless
tiny decimal rounding differences so real problems stand out.
- Using this tool, the team found and fixed mismatches in a large
number of filters (including ATR Trend Reversal, Day-of-Week,
Average-Cross filters, MACD-based filters, DPO cross filters, and
several others) and in several indicators (pivots, AccumDist,
Chaikin Oscillator, Median Oscillator, Zero Crossings, and more).
- Net result: confidence that what you see in GSB matches what the
same strategy would do on the reference platforms has gone up
significantly across this set of filters and indicators.
- One specific fix worth calling out: a day-range calculation used by
the AccumDistDR0 indicator was producing the wrong value
specifically on Kase bars (and other synthetic, non-time-based
bars); this is now fixed and verified to match.
- Day-bar reliability was also tightened during this period: GSB is
more careful about when it considers daily data trustworthy enough
to use, with clearer guidance shown when day-resolution data is too
sparse or unavailable for a given price file.
2. NEW FEATURE: OPENING RANGE BREAKOUT (ORB)
------------------------------------------------------------
A new constrained Opening Range Breakout indicator, filter, and
protective stop have been added, available consistently on every
supported trading platform (GSB, TradeStation, MultiCharts, and
NinjaTrader).
3. NINJATRADER SUPPORT EXPANDING
------------------------------------------------------------
Support for NinjaTrader 8 continues to grow:
- Many more indicators have been ported and verified to match,
including price/value/volume indicators, AccumDist variants,
calendar-day-based indicators, and others.
- 13 additional filters (including time-of-day and day-of-week based
filters) are now available and verified-ready on NinjaTrader.
- Numerous compatibility and compile issues specific to NinjaTrader
were tracked down and fixed so these ports run reliably.
4. "WHAT'S NEW" PANEL (the feature you're reading about right now)
------------------------------------------------------------
A new in-app "What's New" panel was added under the Help menu:
- Automatically pops up after an update, with an option to silence it.
- Available in 4 languages.
- Highlights new settings directly where you'd find them for
example, a star icon next to new options in the settings grid, and
a "New only" filter so you can quickly see just what changed.
5. SETTINGS MANAGEMENT IMPROVEMENTS
------------------------------------------------------------
- A new, easier-to-use editor for picking which Entry Modes,
Normalization Modes, Entry Orders, Exit Modes, Stops, and Targets
are enabled with search, sorting, and platform-aware filtering
(only showing items relevant to your selected platform).
- The same style of editor was extended to the Filters and Indicators
enabled-lists.
6. DATA HANDLING FIXES
------------------------------------------------------------
- Fixed a bug where automatically generated daily data files could be
saved in a date format that GSB couldn't read back in. Daily data
files now save in a consistent, readable format and several older
file date formats are now recognized when loading.
- Fixed a price-data session configuration issue affecting certain
24-hour futures data feeds (extra weekend bars being included).
7. RESULTS GRID: HIDE EMPTY COLUMNS
------------------------------------------------------------
- Optimization results grids now automatically hide columns that have
no data for the current run (for example, Out-of-Sample columns
when OOS wasn't used), so the grid is less cluttered by default.
- A redesigned Columns submenu groups columns by period (Full,
Sample, Validation, Latest-Days, Out-of-Sample, Current, and so on)
with tooltips, and lets you bring any hidden column back with a
single click if you need it.
8. GENERAL POLISH
------------------------------------------------------------
- Cleaned up and unified icon colors and styles across the app.
- Fixed a tooltip on the Price Data screen that could overlap its own
dropdown or fail to reappear.
- Various small reliability fixes to version tracking, so settings
saved under an older version continue to behave correctly when
reloaded.
Thanks received (6):
+1 Stone at 2026-08-25 02:52:58 +1 Ketil at 2026-06-27 10:16:08 +1 JozefSusko at 2026-06-26 06:37:35 +1 cotila1 at 2026-06-26 05:40:05 +1 Bruce at 2026-06-26 03:26:35 +1 Carl at 2026-06-26 01:20:21
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Just to clarify, thats whats in the next release, but I have not released it to all GSB users. Its a major update and so needs more testing
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
75.63 released to beta testers. Anyone who wants a copy to test, email me.
this is the speed difference
  
|
|
|
thowoc213
Junior Member
Posts: 40
Registered: 22-8-2021
Location: Stockholm, Sweden
Member Is Offline
|
|
75.630
The speed improvement is impressive!
I run standard ES 30 min from 2008 and GSB built 2500 unique systems in less than 90 seconds!
Thomas
Thanks received (1):
+1 admin at 2026-07-04 07:12:30
|
|
|
JozefSusko
Junior Member
Posts: 26
Registered: 24-6-2020
Member Is Offline
|
|
75.630
I've done the first tests and I can say it's a massive speedup compared to the previous version. It's faster both during strategy building and
Walk-Forward (WF). I'll keep testing and if any issues come up, I'll definitely let you know.
Peter you've done a really great job.
Thanks received (1):
+1 admin at 2026-07-04 07:12:10
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Thanks for feedback. if anyone has issues, send me an email and under help, do a support ticket
|
|
|
JozefSusko
Junior Member
Posts: 26
Registered: 24-6-2020
Member Is Offline
|
|
75.63
It's a noticeable speed improvement!
1st Image: 70k systems < 40 min.!!!
2nd Image: First workers found in about 1 min. The whole building process of 10k systems took < 8 min.
4th Image: WF workers found < 2 min. The whole WF process for 5 systems took only 12 min.
This is an incredible improvement!!
No issues so far
   
Thanks received (1):
+1 Ketil at 2026-07-05 11:23:10
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
GSB 75.940 RELEASED TO ALL USERS via GSB Resource manager
A major update. Much faster, more polished
Unless you have received GSB from me via email as a beta tester, you will not have beta tester features.
The permissions file for this is called mode or mode.txt in the c:\gsb folder
mode.txt will be in the private forum https://trademaid.info/forum/viewthread.php?tid=249#pid4397
What's New in GSB (since build 70.65)
Broader Data & Bar Type Support
- Day, Week, and Month bars now supported as primary chart data not just intraday.
- Complex price-file naming patterns (e.g. Kase bars with custom interval/session suffixes) now load correctly.
- 11 new built-in trading sessions for major US futures markets (CME Equity Index, Interest Rates, COMEX Metals, NYMEX/ICE Energy, CBOT Agriculture,
CME FX ETH and RTH variants), with tooltips showing exact trading periods.
- Default session for new price files changed from 24/7 → 24/5 (24/7 can't be replicated in TradeStation this was a common source of
GSB/TradeStation mismatches).
- Day-resolution bars can now be automatically "borrowed" from a companion Minute/1-Day series for Kase, Renko, etc.
- Price Data form gained an "Add Day Data" button with context-aware guidance.
- OpenD/HighD/LowD/CloseD and day-bar calc now work across all chart types (Kase, Renko, Tick, Volume previously Minute/Second only).
- Indicators/filters/orders needing day data are now auto-excluded with a clear explanation when unavailable, instead of failing silently.
New Indicators, Filters & Stops
- AtrTrendReversal filter.
- 16 new momentum/oscillator filters: 5 "AverageFcCross" signals (RSI, CCI, Momentum, ROC, Stochastic) + 11 IACF signals (Chaikin Oscillator, Money
Flow Index, TRIX, DPO, Ease of Movement, A/D, %R, MACD Diff, Efficiency Ratio, Linear Reg Slope, Trend Bias).
- Constrained Opening Range (ORB) indicator + filter + protective stop, later made fully configurable (5480 min, bindable to secondary price series),
with 60-min data support and correct handling on early/half trading days.
NinjaTrader 8 Support
- Full NinjaScript generation for all stop types, profit targets, secondary filters, entry modes (matching EasyLanguage behavior, incl. break-even
stops).
- 7 more indicators ported (Close, High, Low, Median, Open, TypicalPrice, WeightedClose) plus calendar-day and warm-up fixes.
- 13 more filters enabled NT8 filter support now at 17 total.
Strategy Execution Fidelity (vs. TradeStation/NinjaTrader/MultiCharts)
- Same-bar re-entry after a stop/target/close exit now suppressed consistently across all platforms.
- Intrabar fills (gap-opens, entry-bar stop-outs, break-even arming) now follow TradeStation's execution model more faithfully.
- Close-of-bar entry/exit/reversal ordering now matches TradeStation.
- Multiple secondary-data-series (Data2+) indicator mismatches fixed across several releases.
- New MatchCheck workspace lets you directly compare GSB's own backtest against generated TradeStation output, with worst-first mismatch reports.
- Multi-series warm-up length (MaxBarsBack) now matches TradeStation.
- Exported EasyLanguage/MultiCharts scripts no longer fail to compile from a stray trailing comma.
Optimization Engine & Performance
- Optimization can now run multiple parallel search "islands" on multi-core machines (/optimizers CLI option), sharing one worker pool/cache.
- Fixed indicator-cache "thrashing" that was silently slowing runs under memory pressure.
- Faster core calculations (StdDev, Average FC, Median FC, Highest/Lowest FC) and reduced memory fragmentation on long Workstation-GC runs.
- Optimization now explains why a component was auto-excluded instead of a generic failure.
- Up-front validation of DataN selections against loaded price files.
Application Log Viewer
- New Application Log window (Help menu) live filterable feed (severity, text search, date range), pause/clear/copy, plus loading saved logs from
previous sessions including Manager/Worker/Admin processes.
Results Grid & Workflow
- Results grid column visibility can now be saved across restarts.
- Grouped "hide empty columns" option organized by metric group.
- Indicators/Filters collection editors gained search, multi-criteria sort/filter, and export/import as .gsboptset.
- Tester Settings shows a clearer parameter-count breakdown before running.
"What's New" / Discoverability
- GSB gained its own What's New window (Help menu), auto-shown after updates, in English/German/Italian/Spanish.
- Recently-added settings/menus/buttons can show a temporary "New" highlight (configurable days, default 180) plus a "New only" filter.
- Saved systems retain the compatibility rules of the build they were created with, so reloading old systems doesn't silently change behavior.
Sharing & Import/Export
- Export/Import System Building Settings (File menu) bundles optimization settings, contracts, sessions, price data, macros into one portable
.share.7z file, with conflict detection on import.
Localization
- GSB now available in German, Italian, and Spanish alongside English.
Thanks received (3):
+1 cotila1 at 2026-07-24 09:49:24 +1 JozefSusko at 2026-07-09 03:28:19 +1 Ketil at 2026-07-09 02:39:52
|
|
|
rws
Member
Posts: 115
Registered: 12-6-2017
Member Is Offline
Mood: No Mood
|
|
Best update ever. Everything runs faster, and the UI is cleaner. What I really like is that the statistics section now lets you see much more clearly
what works and what doesn't; there's far less guesswork involved. That feature isn't found in other genetic builders.
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
76.16 updated for all users. New eld's and bug fixes .
You need to update managers and workers.
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Under testing
New build of GSB Resource manager 20260717
Fix for new build of GSB workers getting terminated quickly when they are idle and should be left alone
Fix for false Anti virus messages if CPU temperature measurements are being done.
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Next release has bug fixes. Nearly all known bugs are resolved. Likely release next week
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
To be released for testing next week.
We now export eld, and pla. Should be no need to ever import the eld/pla files that come with GSB
(This is technology from batch processor making its way into GSB
Thanks received (1):
+1 cotila1 at 2026-08-07 12:37:09
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
main task this week was get all ninja trader indicators to match GSB. GSB NT is close to release.
anyone who wants a beta build to test things like gsb export of eld and .pla files, let me know
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
76.35 released. lots of bug fixes and NT. Installer for new users or new installs is busted. Working on this tomorrow.
Users with GSB Resource manager can auto download the files. Managers and workers need update, and new eld has to be loaded unless you use .eld / .pla
export mode
|
|
|
REMO755
Member
Posts: 183
Registered: 11-4-2021
Member Is Offline
|
|
Quote: Originally posted by admin  | 76.35 released. lots of bug fixes and NT. Installer for new users or new installs is busted. Working on this tomorrow.
Users with GSB Resource manager can auto download the files. Managers and workers need update, and new eld has to be loaded unless you use .eld / .pla
export mode |
Hello,
I've been waiting four years for something that works on NT8.
Where can I download it? You said you'd email it to me?
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Remo, I had files but installer was busted. have it going. will send you email now
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Next week bug fixes. To export eld from GSB, some of the required files were not in the update package.
Had reasonable success with 90 min day trade systems on NQ
Slow progress this week on AI trading app. At a guess at new highs trading (I have stock in the same account so that affects the PL)
more refinements and bug fixes on batch processor being working on.
Thanks received (1):
+1 cotila1 at 2026-08-28 03:18:51
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
77.52 RELEASED. This version will not talk to older managers, so workers and managers have to be updated via GSB Resource manager.
It has many bug fixes. To use the free GSB cloud you must update.
Thanks received (1):
+1 cotila1 at 2026-09-12 02:17:58
|
|
|
Eugene
Junior Member
Posts: 3
Registered: 26-5-2017
Member Is Offline
Mood: No Mood
|
|
Great migration to modern .Net works! Have you also upgraded to the newer C#?
As i recall, that would make custom/user indicators implementation feasible.
|
|
|
Eugene
Junior Member
Posts: 3
Registered: 26-5-2017
Member Is Offline
Mood: No Mood
|
|
Also: any plans to allow more flexibility with creation of custom fitness functions?
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Eugene
we already have a lot of options for fitness, but regardless I will never use anything but np*at to build systems.
This gives a balance between NP and aver trade. NP only will give trades that have too low profit after slippage and commission
to high emphasis on ave trade gives too few trades, which leads to high system degradation out of sample.
But as for final system fitness, what you as the human choose, we have lots of options. I look at np/dd, and how linear the equity curve is (pearsons
to a straight line)
and look at the walk forward curves
GSB on developement works up to dotnet 10, but there was zero speed advantage, and the licensing isnt compatible.
The good news is recent build of the last few months is many times faster than the old build.
My focus now is on the new product which is AI system building. Hope to have beta build in a month roughly, but to hard to tell.
The AI gives you massive flexability in what you want to do.
|
|
|
Eugene
Junior Member
Posts: 3
Registered: 26-5-2017
Member Is Offline
Mood: No Mood
|
|
Question on C# (not the .net) was actually more about ability to use Span (which is much faster...probably this was the reason for the recent
speedup).
As well as language facilities allowing to use custom user-defined/calculated indicators.
As for Fitness Functions for building: genetic search is directed by the gradient of components.
NP*AT = NP^2/ TT.
Depending on the instrument's range (in %), NP might or might not 'counter-balance' gradient of TT.
(E.g. say, gene 1 got NP=100 and TT=400 => np*at=25;
say, gene 2 got 10 more PROFITABLE trades.
For FX(say EURUSD) - with daily range~0.5% and profit/trade of,say, 4 ATR - NP will become 100+10*4*(1.2*0.005)=100.24 (or +0.24%), BUT trades
'gradient' is 2.5% (10/400)!
so, NP*AT=(100.24*100.24)/410=24.5 - > it went down (!) despite 10 more profitable trades!
For instruments with much higher ranges (NQ, ES, CL, individual stocks), this would turn differently, and is why "it is easy to build good systems
with GSB for ES or NQ".
But even with such instruments NP*AT still generally 'finds'/prefers systems with smaller number of trades (read: less robust AND requiring a lot of
tinkering with params afterwards to get to decent trade numbers).
Also, NOT normalizing metrics to vol brings another 'artifact' that hinders search/ discovery of real alpha: search will overfit to periods with vol
explosions (look at SI or GC or CL).
It is better to use ratios (like PF) and prefer more trades, not less, managing 'gradients' of metrics with tahn, log, pow, etc.
It would be great to have either more in-built 'facilities' for doing the above in GSB, or an API for doing it ourselves.
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@Eugene
Thanks I think your point about the search gradient is important.
In practice NP*AT and NP*PF can behave fairly similarly because both are still strongly influenced by raw NP, so they can inherit the same
volatility/regime bias and can also push the genetic search toward fewer trades than we really want.
I can add a couple of alternative fitness functions to GSB so we can compare them directly against NP*AT and NP*PF:
- `sqrt(TT) * ln(PF)` rewards more trades, but with diminishing returns, while avoiding raw NP as the main driver.
- SQN-style fitness preferably using volatility-normalised trade returns, so the search rewards edge, consistency and sample size rather than simply
dollar profit.
I also like your suggestion of allowing the gradients themselves to be shaped. I can add support for transforms such as `tanh`, `log`, `sqrt/pow`,
etc., so metrics such as PF, trade count, average trade and expectancy can be bounded or weighted without one component dominating the search.
Longer term, I agree that volatility-normalising the relevant metrics is probably important. Otherwise the optimizer can mistake high-volatility
periods for alpha, particularly in markets such as GC, SI, CL and NQ.
So rather than committing to one replacement for NP*AT, I think the best approach is to expose these as alternative fitness choices and test which
ones consistently produce more robust systems across different instruments and regimes.
I have added %Profitable*AT*Trades at the very top of the list.
SQN near the bottom, just below SharpeRatio to be released in the next build
we now have all these and combinations of in fitness
My quick tests showed no difference in results between np*at and sqn and %Profitable*AT*Trades, but because of random seed, many identical tests need
to be done to clearly see small edges.
I did only one test of each. Next time im in system building mode, I will try more tests
I will release update when I do the next release. Not sure when but if your in a hurry for it , let me know
%Profitable*AT*Trades
%Profitable*AvgWin/(AvgWin+AvgLoss)
%Profitable*PF*Trades
%Profitable*RatioAvgWinAvgLoss
ASTAB_C
AvgDistanceBetweenNewHighs
AvgOfTopFiveDrawdowns
AvgTrade
Drawdown
LSR
LSR+1
MAE*-1*10000
MedianTrade
MFE
Mult10000
NegExcursion
NetProfit
NetProfitOverAvgOfTopFiveDrawdowns
NetProfitOverDrawdown
NP / -DD - W (TS fitness)
NP*%Profitable*AvgWin/(AvgWin/AvgLoss)
NP*%Profitable*RatioAvgWinAvgLoss
NP*AT*MFE*NPLateEntry
NP/DD*(Pearson-0.5)*PF*SR
NPAIPossibleLateEntry
NPLateEntry
Pearson
Pearson-0.5
PearsonByTradeDate
PercentProfitable
PFAIPossibleLateEntry
PFLateEntry
ProfitFactor
PROM
RatioAvgWinAvgLoss
RSTAB_C
SharpeRatio
SQN
SumOfAreasOfDrawdown
TmPlusMinus
TradesCount
TradeStdev
VBasePlusMinusLong-25d
VBasePlusMinusMedium-10d
VBasePlusMinusShort-5d
WalkForwardEfficiency
|
|
|
admin
Super Administrator
Posts: 5122
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@Eugene
Thanks I think your point about the search gradient is important.
In practice NP*AT and NP*PF can behave fairly similarly because both are still strongly influenced by raw NP, so they can inherit the same
volatility/regime bias and can also push the genetic search toward fewer trades than we really want.
I can add a couple of alternative fitness functions to GSB so we can compare them directly against NP*AT and NP*PF:
- `sqrt(TT) * ln(PF)` rewards more trades, but with diminishing returns, while avoiding raw NP as the main driver.
- SQN-style fitness preferably using volatility-normalised trade returns, so the search rewards edge, consistency and sample size rather than simply
dollar profit.
I also like your suggestion of allowing the gradients themselves to be shaped. I can add support for transforms such as `tanh`, `log`, `sqrt/pow`,
etc., so metrics such as PF, trade count, average trade and expectancy can be bounded or weighted without one component dominating the search.
Longer term, I agree that volatility-normalising the relevant metrics is probably important. Otherwise the optimizer can mistake high-volatility
periods for alpha, particularly in markets such as GC, SI, CL and NQ.
So rather than committing to one replacement for NP*AT, I think the best approach is to expose these as alternative fitness choices and test which
ones consistently produce more robust systems across different instruments and regimes.
I have added %Profitable*AT*Trades at the very top of the list.
SQN near the bottom, just below SharpeRatio to be released in the next build
we now have all these and combinations of in fitness
My quick tests showed no difference in results between np*at and sqn and %Profitable*AT*Trades, but because of random seed, many identical tests need
to be done to clearly see small edges.
I did only one test of each. Next time im in system building mode, I will try more tests
I will release update when I do the next release. Not sure when but if your in a hurry for it , let me know
%Profitable*AT*Trades
%Profitable*AvgWin/(AvgWin+AvgLoss)
%Profitable*PF*Trades
%Profitable*RatioAvgWinAvgLoss
ASTAB_C
AvgDistanceBetweenNewHighs
AvgOfTopFiveDrawdowns
AvgTrade
Drawdown
LSR
LSR+1
MAE*-1*10000
MedianTrade
MFE
Mult10000
NegExcursion
NetProfit
NetProfitOverAvgOfTopFiveDrawdowns
NetProfitOverDrawdown
NP / -DD - W (TS fitness)
NP*%Profitable*AvgWin/(AvgWin/AvgLoss)
NP*%Profitable*RatioAvgWinAvgLoss
NP*AT*MFE*NPLateEntry
NP/DD*(Pearson-0.5)*PF*SR
NPAIPossibleLateEntry
NPLateEntry
Pearson
Pearson-0.5
PearsonByTradeDate
PercentProfitable
PFAIPossibleLateEntry
PFLateEntry
ProfitFactor
PROM
RatioAvgWinAvgLoss
RSTAB_C
SharpeRatio
SQN
SumOfAreasOfDrawdown
TmPlusMinus
TradesCount
TradeStdev
VBasePlusMinusLong-25d
VBasePlusMinusMedium-10d
VBasePlusMinusShort-5d
WalkForwardEfficiency
|
|
|
| Pages:
1
..
53
54
55
56 |
![[*]](images/davis/default_icon.gif){kind=icon}
