Commit graph

000c45fdbb feat: print a self-verifying config line per chart at startup main AnimateDread 2026-07-29 20:22:22 -04:00
70fed28015 docs: correct the conv-pool rationale against the reference contract AnimateDread 2026-07-29 19:53:47 -04:00
70cdec2717 fix(ai): drop the conv pooling stage - it reduced across filters, not time AnimateDread 2026-07-29 19:28:44 -04:00
f2ec1edf84 feat(ai): logit-adjusted loss, replacing oversampling and the post-hoc prior AnimateDread 2026-07-29 19:05:14 -04:00
2695a961c4 refactor(perf): pin CPU threads per network, drop the TargetCPULoad input AnimateDread 2026-07-29 15:33:37 -04:00
a2d4a7b218 fix: tag AI signal names with the config fingerprint AnimateDread 2026-07-29 15:28:53 -04:00
d4dea8d6f2 fix: close five weight-affecting inputs missing from the model fingerprint AnimateDread 2026-07-29 15:21:49 -04:00
8ae27bec08 fix: name AI signals by dense depth so concurrent charts are separable AnimateDread 2026-07-29 15:11:30 -04:00
2ad9fdd531 fix: refuse to start when another chart owns the same model files AnimateDread 2026-07-29 14:49:54 -04:00
54519e9917 fix(perf): treat TargetCPULoad as a machine budget split across charts AnimateDread 2026-07-29 14:06:35 -04:00
692cb0eeaa refactor(ai): derive the dense taper's shape, not just its first layer AnimateDread 2026-07-29 14:03:42 -04:00
af209997fc refactor(ai): derive the first dense layer's width instead of asking for it AnimateDread 2026-07-29 13:01:16 -04:00
77184cdb11 docs: record the .nnw architecture-persistence root cause and the batch-norm gap AnimateDread 2026-07-29 12:38:40 -04:00
65dc1bddb4 fix(ai): freeze batch-norm statistics when comparing two forward passes AnimateDread 2026-07-29 12:37:50 -04:00
30206cbabc feat(ai): batch normalization between dense layers AnimateDread 2026-07-29 12:34:29 -04:00
1aa7df9096 fix: stop a .nnw from pinning a superseded architecture AnimateDread 2026-07-29 12:00:40 -04:00
4cb888e1b1 fix: clear stale signal arrows when a fresh model starts at era 0 AnimateDread 2026-07-29 10:13:01 -04:00
cc625c827e fix(training): escape the recall-gate catch-22 that let runs decay unchecked AnimateDread 2026-07-29 07:51:08 -04:00
41341f44c1 fix(panel): show the metric that actually gates deployment AnimateDread 2026-07-29 07:15:45 -04:00
26ac479bae docs: refactor + audit notes; fix malformed comment banner AnimateDread 2026-07-29 00:46:01 -04:00
2de93539d4 refactor: split CExpertSignalAIBase implementation by responsibility AnimateDread 2026-07-29 00:42:45 -04:00
d9a4f91717 refactor: compose topologies from named stages; drop dead code AnimateDread 2026-07-29 00:38:05 -04:00
41d6a63d92 fix: make sidecar writes atomic; extract shared AtomicFile helper AnimateDread 2026-07-29 00:31:29 -04:00
4f28165cd3 fix: remove broken DFA optimizer, use plain gradient descent AnimateDread 2026-07-29 00:03:54 -04:00
2fc587efa0 fix(optimizer): remove broken DFA optimizer AnimateDread 2026-07-29 00:02:57 -04:00
165898119a fix: add null pointer checks for neuron casts in backpropagation AnimateDread 2026-07-28 21:28:55 -04:00
5971f14a73 fix: handle OCL neuron types in gradient clipping and backprop loops AnimateDread 2026-07-28 20:46:27 -04:00
1b80e73e34 perf: replace dynamic_cast with switch-type casts in optimizer methods AnimateDread 2026-07-28 19:59:10 -04:00
d616cfef1b refactor(Network): replace type switches with dynamic_cast AnimateDread 2026-07-28 18:13:21 -04:00
7d3dde9d31 refactor: standardize updateInputWeights signatures across neuron classes AnimateDread 2026-07-28 17:58:14 -04:00
450017d6fd fix(Network): pass SourceObject by reference in updateInputWeights AnimateDread 2026-07-28 17:52:13 -04:00
30c0aafff8 feat(Network): add DFA training and optimizer snapshot support AnimateDread 2026-07-28 17:42:12 -04:00
d988c8829a refactor(inputs): switch default training optimizer to DFA AnimateDread 2026-07-28 15:05:56 -04:00
fd5dabfed8 refactor(AI): rename gradient parameter to localGradient in DfaFeedbackSignal AnimateDread 2026-07-28 15:02:57 -04:00
c32e104e8f feat(opencl): add feedback alignment support to weight update kernels AnimateDread 2026-07-28 15:01:40 -04:00
a78dc37a0a docs: standardize market description HTML and translations AnimateDread 2026-07-28 13:07:30 -04:00
a303f5b86c refactor: merge AI topology preset into AI_CHOICE enum AnimateDread 2026-07-28 12:02:58 -04:00
52d8cee308 fix: use TopologyPreset in DB config fingerprint AnimateDread 2026-07-28 11:51:33 -04:00
e4f88d7934 feat: replace HiddenLayersCount with AI_TOPOLOGY_PRESET for architecture-aware topology presets AnimateDread 2026-07-28 11:47:29 -04:00
941c88a20f refactor(hyperparams): soften labels, shrink network, tune priors and gamma AnimateDread 2026-07-28 10:49:53 -04:00
031f147b2a feat: change default AI type to HYBRID and bump version to 3.2 AnimateDread 2026-07-27 22:30:06 -04:00
2285ddf697 refactor: replace AIType with discrete EnablePAI/CONV/LSTM/HYBRID flags AnimateDread 2026-07-27 22:18:50 -04:00
e043e565eb feat: implement hybrid AI signal with CNN-LSTM architecture and add pooling parameters AnimateDread 2026-07-27 22:08:55 -04:00
19c25d74d2 docs: polish English market description for clarity and consistency AnimateDread 2026-07-27 16:35:15 -04:00
48abb89e6a fix: skip DirectML DLL in tester, add NaN guards, improve chart cleanup AnimateDread 2026-07-27 15:52:39 -04:00
13562221f4 refactor: use LOGIT_PRIOR_OFF as default for AILogitPriorStrength AnimateDread 2026-07-27 11:53:47 -04:00
3a37b9115e fix: correct inference-only new-bar detection and add stop validation AnimateDread 2026-07-27 11:51:45 -04:00
146881c507 fix: allow inference-only tester to trade with loaded model when trainingComplete is false AnimateDread 2026-07-27 11:28:23 -04:00
b1dd61d0da feat: add signals visibility toggle and tester rejection tracing AnimateDread 2026-07-27 11:13:19 -04:00
061c3d1c8b feat: suppress noisy performance warnings unless VerboseMode AnimateDread 2026-07-27 10:58:29 -04:00
53599d5883 Merge branch 'main' of https://forge.mql5.io/animatedread/Warrior_EA AnimateDread 2026-07-27 09:25:03 -04:00
ff2ece0249 fix: increase min activation deriv floor and seed RNG for fresh topology AnimateDread 2026-07-27 09:24:53 -04:00
9108cdb44b docs: overhaul German market description for clarity and new features Jason Binet Lafontaine 2026-07-26 23:52:28 -04:00
0612eba754 docs(market): enhance MACD and Ichimoku period explanations Jason Binet Lafontaine 2026-07-26 23:33:14 -04:00
a228d1bde7 feat(trade): implement trade safety checks per Article 2555 and resource limits Jason Binet Lafontaine 2026-07-26 23:08:32 -04:00
5da111d61f fix: correct totalTrades variable type to double Jason Binet Lafontaine 2026-07-26 21:15:57 -04:00
9882ea929e feat: add OnTester linear equity optimization with drawdown decimal fix Jason Binet Lafontaine 2026-07-26 21:09:53 -04:00
1e4c54b95a refactor: update classic signal presets and disable close vote by default Jason Binet Lafontaine 2026-07-26 18:48:34 -04:00
a29e6677b0 docs: update market descriptions for MACD/Ichimoku and revised voting thresholds Jason Binet Lafontaine 2026-07-26 18:44:07 -04:00
b2069bcee4 feat(signals): add MACD/Ichimoku presets and Vote_Close disabled option Jason Binet Lafontaine 2026-07-26 18:33:12 -04:00
8d4fe088b8 refactor: unify AI and classic vote parameters to Min_Vote_Open/Close Jason Binet Lafontaine 2026-07-26 17:27:51 -04:00
a17f8f1e15 fix(signal): snapshot alternation gate to prevent premature consumption on discarded votes Jason Binet Lafontaine 2026-07-26 17:09:13 -04:00
08aefbc736 :752 — the window average is now computed after folding in this call's own result, so what's returned always includes the current tick. :106 — window key changed from the 0-59 sec field to a full datetime. Fixing #1 alone would have replaced "always 0" with "cumulative average of every bar since startup", since the window still never rolled over. :746 — restored result /= number. number was being counted and then never used, leaving a raw sum where the base class averages. MA's 60 + RSI's 100 = 160 tripped the ±100 range check and got zeroed — it discarded precisely the strongest agreed setups. Jason Binet Lafontaine 2026-07-26 15:31:09 -04:00
58bdac1328 fix: handle legacy neuron classes in BlendWeightsFrom to avoid UB AnimateDread 2026-07-26 14:45:08 -04:00
3c8f41e6ab fix: preserve historical arrows on rescan and remove temporary throughput diagnostics AnimateDread 2026-07-26 14:08:59 -04:00
0233664d29 feat: add raw argmax tally to rescan logs, raise train time budget to 200ms AnimateDread 2026-07-26 13:58:05 -04:00
f193429299 refactor: make RescanPending() public and update comments for deferred rescan logic AnimateDread 2026-07-26 12:55:31 -04:00
1bef4fea08 refactor(expert): split RescanChartSignals into async start/advance methods AnimateDread 2026-07-26 12:52:56 -04:00
15226b64df feat: add manual rescan of chart signal arrows from deployed model AnimateDread 2026-07-26 12:36:56 -04:00
5247c34fe9 fix: add error logging for buffer failures and reject trades on invalid stop loss AnimateDread 2026-07-26 12:12:14 -04:00
c5f3d56542 chore: move and rename Market_Description.html to Market Descriptions/ subdirectory AnimateDread 2026-07-26 12:11:56 -04:00
fc0b244aa0 feat: implement deferred chart arrow restore to avoid frozen OnInit AnimateDread 2026-07-26 11:17:55 -04:00
2cf68e2e87 fix: skip weight save on inference-only runs and detect stale cache AnimateDread 2026-07-26 10:59:46 -04:00
b7ef83f8c6 fix(ai,expert): add FILE_SHARE flags to avoid 5004 errors on deployed models AnimateDread 2026-07-26 10:46:19 -04:00
6e47019be5 refactor: add exponential backoff retry logic and bar-close autosave AnimateDread 2026-07-26 10:27:38 -04:00
377aa6bbc4 fix: handle transient file locking during concurrent model saves and loads AnimateDread 2026-07-26 10:15:56 -04:00
2b666f6744 refactor: compress whitespace and divs in Market_Description.html AnimateDread 2026-07-26 09:19:38 -04:00
c30a10f7b9 fix(ExpertSignal): convert TP from ATR-relative to risk-relative to fix zero-trade bug AnimateDread 2026-07-25 22:33:45 -04:00
8ae14c60a8 fix: protect TargetCPULoad for Market build and update description AnimateDread 2026-07-25 21:32:15 -04:00
d6bbf26785 fix(ExpertSignalAIBase): add primary occurrence flag to correct IS accuracy under oversampling AnimateDread 2026-07-25 19:34:24 -04:00
69307ff121 feat: implement manual deploy and retrain in ExpertSignalAIBase AnimateDread 2026-07-25 16:39:11 -04:00
c7e533a880 feat(ExpertSignalAIBase): add plateau ladder escalation and legacy MinWR shim AnimateDread 2026-07-25 15:55:56 -04:00
d5575db339 fix: suppress repeated OpenCL/DirectML probe messages across multiple CNet instances AnimateDread 2026-07-25 13:03:45 -04:00
582e1dec59 fix: correct CLayer::CreateElement signature to prevent model load failures AnimateDread 2026-07-25 12:02:38 -04:00
b09c3a5f4e feat(network): add quiet parameter to CNet::Load for graceful miss handling AnimateDread 2026-07-25 11:05:28 -04:00
f70824b8c2 fix(AI): correct FileMove destination flags to prevent model save in wrong sandbox AnimateDread 2026-07-25 10:40:47 -04:00
703a1f46d6 fix: improve load diagnostics and prevent shutdown crash AnimateDread 2026-07-25 02:01:05 -04:00
5d6f03014e fix: implement atomic file save in CNet::Save to prevent data corruption AnimateDread 2026-07-25 01:21:05 -04:00
5fda52b212 docs: clarify cumulative accuracy as directional win-rate excluding neutral calls AnimateDread 2026-07-25 01:07:21 -04:00
3f0d1557de feat: add compounded accuracy counters and remove fired-hitrate threshold AnimateDread 2026-07-25 00:02:34 -04:00
b9fe2d865f fix: reorder shutdown to ensure chart cleanup before weight persistence AnimateDread 2026-07-24 21:56:53 -04:00
aba963a97d feat: require minimum fired calls for hit-rate display AnimateDread 2026-07-24 21:42:26 -04:00
635cb555c1 fix: force isInitialized=false on weight reset to avoid config mismatch AnimateDread 2026-07-24 21:36:02 -04:00
d7c550c470 feat: add in-memory weight snapshot/restore for mid-run stability AnimateDread 2026-07-24 21:21:47 -04:00
339369fe99 feat: add common parameter to checkpoint save/load for consistent file location AnimateDread 2026-07-24 18:56:20 -04:00
bf82f4a034 fix(AI/Network): atomic checkpoint restore to prevent zero-layer state on failure AnimateDread 2026-07-24 18:37:54 -04:00
4f5003dc78 fix: prevent CLayer memory leak on network load failure AnimateDread 2026-07-24 12:01:09 -04:00
b78bb5d2ec feat: support pure-MQL5 CPU inference for OCL neuron layers without GPU backend AnimateDread 2026-07-24 11:52:19 -04:00
303d46da8b fix: reset training state on failed network load to prevent untrained model deployment AnimateDread 2026-07-23 21:46:38 -04:00
72797ab7ba feat: prevent saving empty network and add logit adjustment calibration AnimateDread 2026-07-23 19:36:34 -04:00