+ (7j4m 2Rt^RIt^RIt^RIt^RIt^RIt^RIt^RI t ^RI t ^RI t ^RI H t H t ^RIt^RIt] P$P']P$P)]P$P+]P$P-]4R444^RIt^RIHtHtHtHtHtHt^RI H!t!^RI"H#t#H$t$^= WFO_MIN_VALID_FOLDS valid folds required) + lambda * mean squared deviation from the pretrained weights (L2 anchor). `--fold-objective robust` keeps the same WFO-only training boundary but subtracts penalties for high fold drawdown, zero P&L/DD, low fold trade coverage, negative folds, and threshold fragility. Per-fold Calmar matches the existing WFO rescore convention in auto_optimize_loop.py (composite_score's 1000% P&L gate is meaningless on ~1-year folds, so Calmar is the per-fold metric). Usage: python3 tools/train_mlp.py --data data/COINBASE_BTCUSD-6H.csv [--hidden 16 8] [--target-k 10] [--phase1-epochs 300] [--es-generations 300] [--es-popsize 32] [--es-sigma 0.05] [--l2 1e-3] [--fold-objective mean|robust] [--es-workers 0] [--seed 42] [--smoke] [--skip-phase2] [--out PATH] Output: strategies/params/mlp/mlp_weights_{ASSET}_{TF}.json N)datetimetimezone..)ACTIVATION_NAME FEATURE_COLS_score_to_signalscalculate_metrics mlp_forwardsave_mlp_artifact)_prepare_features)DEFAULT_WORKER_FRACTIONresolve_worker_countg@Y@gI@皙??c\P!R\PP V44pV'g\ RV 24hVP ^4VP ^43#)z ([A-Z0-9_!]+)-([0-9]+[HDW])\.csvz&Cannot derive ASSET/TF from filename: )rematchospathbasename ValueErrorgroup) data_pathms& :/Users/jameslopez/projects/TradingBot25/tools/train_mlp.pyderive_asset_tfrIsS 4bgg6F6Fy6QRA A)MNN 771:qwwqz !!c\P!V4pVPPP 4PP 4Vn\P !VR,RR7PPR4VR&VR,\P8VR,\P8*,pVPV,PRR7#)timeT)utcNdrop)pdread_csvcolumnsstrlowerstrip to_datetimedt tz_localizeconfig TRAIN_START TRAIN_ENDloc reset_index)rdfmasks& r load_datar3Ps Y B%%'++113BJ6 588DDTJBvJ vJ&,, ,Fv?O?O1O PD 66$< # # # ..rc\P!VR,PP\P44p\ P !\P!W"^,R74pVP\4P4P^4Pp\P!\V4\P4pW!RVRV), V, VRV)%\P!\ V,VR,, 4p\P"!V4(pWpR,\$P&8P,pWg3#)zJtanh-squashed vol-normalised k-bar forward log return. Returns (y, valid).close)prependNư>r)nplogvaluesastypefloat64r#Seriesdiffrolling VOL_WINDOWstdshiftfulllennantanh TARGET_SCALEisnanr, SCORE_START)r1k log_closeret1volfwdyvalids&& r build_targetrQXsr'{))00<=I 99RWWY! = >D ,,z " & & ( . .q 1 8 8C ''#b'266 "C" #A2.!3C!H  s"cDj12A XXa[LE jF... 6 66E 8Orc  aaa!a"^RIo"S"PP4'dRM-S"PPP4'dRMRo!S"P V4\ PPV4SP^,.\V4,^.,p .p .p \\V 4^, 4Fep S"PPW,W^,,4pWS"PP4., p V P!V4Kg S"PP"!V !P%S!4pW#V8P&,W88*P&,pW#V8P&,pV !RS! RV RVP)4 RVP)4 24VV!V"V3R lpV!V4wppV!V4wppS"P*P-VP/4R R R 7pS"PP14p\ P2R^^3wpppp\V4EFpVP54VP74V!V!V4V4pVP94VP;4VP=4pVP?4S"PA4;_uu_4\V4'd\CV!V!V4V44MTp RRR4X VR , 8dT ^ppV Uu.uFpVPDPG4PI4PK4PM4PO4VPPPG4PI4PK4PM4PO43NK ppM"V^, pVV8dV !RV RVR R24M)V^2,^8XgEKV !RV RVR RV R 24EK VRS!RV RX^,RVR\SVP)44R\SVP)44/3# +'giELg;iuupi)Ncudampscpuzphase1: device=z arch=z train_bars=z val_bars=c<SPSV,SPSR7SPSV,SPSR7P^43#))dtypedevice)tensorfloat32 unsqueeze)mask_arrXrYtorchrOs&rtenspretrain..tenssO Qx[ f M Qx[ f MWWXYZ\ \rg{Gz?g-C6?)lr weight_decayr7zphase1: early stop at epoch z (best val MSE z.5f)zphase1: epoch z train=z val=rYarch epochs_run best_val_mse train_barsval_bars)*r_rT is_availablebackendsrU manual_seedr8randomseedshapelistrangerDnnLinearTanhappend Sequentialtor:sumoptimAdam parametersMSELossinftrain zero_gradbackwardstepitemevalno_gradfloatweightdetachrVdoublenumpycopybiasint)#r^rOrPtimeshiddenepochsrn val_startval_endr9relayers_tmodulesilinmodelval_mask train_maskr`XtrytrXvayvaoptloss_fnbest_val best_statepatience since_bestepochloss train_lossval_lossrYr_s#ff&&&&&&&& @@rpretrainrjs //11f!NN..;;==5  dIINN4 GGAJ<$v, &! ,DHG 3t9q= !hhoodgtE{3))" HH   ) , ,V 4E*222e6F5N5NNH)+333J/&v\*..:J9KLLLN# %&\\JHCH~HC ++  5++-$T  JChh G13r11D.Hj(Jv  uSz3'   YY[  ]]__:=c((uWU3Z56 H ho %#+QjH&./%-c::,,.224;;=CCEJJL88??,00299;AACHHJL%- /J !OJX%25'RUVWXY 2:? .wz#.>eHS>R S+. &&$ eai&$c*..*:&;"C $79 99__/s',Q11B-R1 R c \P!VUUu.uF9wr\P!VP4VP4.4NK; upp4#uuppiN)r8 concatenateravel)layersWbs& rflatten_layersrs? >>fUfda2>>1779aggi*@AfU VVUs?A c .^r2\\V4^, 4FkpW^,,W,reWW5V,,PWV4qsWV,, pWW5,qV, pVPWx34Km V#)rS)rqrDreshaperu) thetarerrlin_outn_inrrs && runflatten_layersrs{AACIM"6lDHt A $ % - -e :>?a HD rvvdm$ %%rc\VR4p\V^,\\P!VR,444p\^\ V44pRpRp Rp Rp VFp \V P R^44p \\V P RR444p\V P RR44p\V P RR44pV\RV\, 4\, , pVR8:d V R, p V \RWm, 4V, , p V \R\RV)44, p K W,pW,p W,p W,p \W#\4p\W#\4pRV,\RVR , 4,pR VR V R V R V RVRVRVRV/pRV,RV ,,RV ,,RV ,,V,pVV, V3#)rg? Total TradesMax Drawdown % P&L/DD Ratio Calmar Ratiorrg@@drawdown_penaltypnl_dd_penalty trade_penaltynegative_fold_penaltyfragility_penalty fragile_0_05 fragile_1_0 trade_targetg?g?r$)rmaxrr8ceilrDgetrrROBUST_DRAWDOWN_TARGETrrROBUST_FRAGILITY_TIGHT_MARGINROBUST_FRAGILITY_WIDE_MARGIN)r fold_metricsrrmin_fold_tradesbasern_foldsrrrneg_fold_penaltymetricstradesdrawdownpnl_ddcalmar fragile_tight fragile_wider penaltiespenaltys&&&&& rrobust_objective_scorers  4D*C#8M0N,OPL!S&'GNMW[[34uW[[)93?@Aw{{>378w{{>59:CX0F%FGJ``` S= c !NS,"78<GG CSvg%677 NM-fB_`M,VA]^L -S,:N1OO ,.!1. |  I        ! ! "     '>9 $$rc RVRR.,P4RVR,P4RVR,PP\P4RVR,PP\P4RVR,RVR \ V4R \P !V\PR 7R \ V4R VRVRVRV/ #)base_dfrr5r close_arrhigh_arrhighstoch_peak_arrr^retheta0_wrfoldsrmin_valid_foldsl2fold_objective)r )rr:r;r8r<rpr) r1r^rerrrrrrs &&&&&&&&&rmake_fitness_contextrs 2vw'(--/F"R[''..rzz:BvJ%%,,RZZ8!D' QT BJJxrzz:e?? b. rcZ\P!V\PR7pVR,p\V4p\ VRVVR,4p\P !WR\ ,.R#O.R$O4p\VR,V4pVR,P4pWgR&VR,VR &R V^,R V^,R V^,R RRRR\RV^,4/p\VVVR,VR,VR,4p..^rp VR,p VR,p VR,FwrpVPW8V V8*,,p\V4^ 8d-V PR^RR%RRRR/4V PR4Ka\VV R7pV PV4VR,R%8:gVR,V 8dV PR4KV PVR,4V ^, p K RV /pVR,pVR8Xd4\V V VVV 4wppVPV4\!V R4VR &MWR!,8dRpM \!V V4pVR",\#\P$!VRVV, ^,44,pV)V,WV3#)&rrNrerr^ractivation_scorerr!i_long_entry_activation_threshold i_long_exit_activation_threshold-i_long_exit_activation_confirmation_thresholdi_use_long_exit_confirmationri_use_long_entry_confirmationFi_trailing_stop_thresholdrrrrrrrrr) min_trades valid_foldsrrobustrbase_fold_scorerr) rrr) @r r g>@r)r8rr<rDrclipTHRESHOLD_SCALEr rrrr/rurrupdaterrr)rcontextrn_wrrrdparamsrrrPrr_ oos_startoos_endslr diagnosticsrscoreranchors&& revaluate_thetars JJuBJJ /Ez"H h-C eDSk76? ;F d o%%#J v .F !A" #AfI+Z]*JqM7A&'#Sjm%< F   !    A$&r15G G E/0O!(!1!1g UUE&5G+;< = r7R<   NE!/6F!M N NN3  #B?CG$ > "e +w~/F/X NN3  NN7>2 3 QJE"2!%(K-.N!1       y 9%)=gz)R %& *+ +$Wn= T]U277E$3K(,Bq+H#IJ JF 8f g; >>rc VsR#r)_FITNESS_CONTEXT)r s&rinit_fitness_workerrYsrc"\V\4#r)rr)rs&revaluate_theta_workerr^s %!1 22rc Ra a \WW#WEWgV4 o V V 3Rlo S S n/S nS #)c8<\VS4wrr4VSnWV3#r)rlast_diagnostics)rfitrrrr fitnesss& rr"make_fitness..fitnessis'0>ug0N-j#. Z''r)rr r ) r1r^rerrrrrrr r"s &&&&&&&&&@@r make_fitnessr$bs4" tuG ( GO!G Nrc aaaaaaaa^RIpVPWRVRVRR/4o\^\S44oR\P RVRRRRR /R S/oVV3R loVVVVV3R lp S^8XdOS!R 4V3Rlp \ S4F/p V !W4SP4'gK#S!RV 24S# S#\SRR4p V f \R4hS!RS R24\P!R4p \PPSV \V 3R7;_uu_4o\ S4F3p V !V V3Rl4SP4'gK(S!RV 24M RRR4S# +'giS#;i \ \"3d)pS!RT R24\%SYTSTS^R7uRp?#Rp?ii;i)rSNpopsizernverboser!rrthrrworkersc <Vwr4rVVPV4VSR,8d<RVR\P!V4P4RVRVR\ V4RS/oR#R#)r!rrr(rr)N)rur8rrdict) solutionresultfitsfrr(rbestr)s &&& robserve_solution!run_cma..observe_solution|sd'-$C A tE{?qH-2247stK07 D rc<S P4Uu.uF)p\P!V\PR7NK+ ppV!V4p.p\ W44FwrgS !WgV4K S P W54V^ ,^8XgVS ^, 8XdJS !RT RS RS R,R RS R,;'g.Uu.uFp\ V^4NK up 24R#R#uupiuupi) rz phase2: gen /z best_fitness=r!.4f fold_calmars=rN)askr8rr<ziptellround)gen evaluators solutionsresultsr.r,r-cr0es generationsr9r1s&& rrun_generationrun_cma..run_generations>@ffhGhRZZ4h GI& #I 7 H Xt 4!8   8q=C;?2 ,se1[MU C?PQ 7;I7L7L"7L N7Lq!7L NOQ R3 H!Os /C$C)z1phase2: evaluating CMA-ES population sequentiallyc <.pVF6pS!V4wr4pVPW4V\\SR/4434K8 V#)r )rur+getattr)r>outr,r/rr(r"s& rsequential_evaluator%run_cma..sequential_evaluatorsPC%")("3C A 2Db!IJLM&Jrz phase2: CMA-ES converged at gen r z*Parallel CMA-ES requires a fitness contextz*phase2: evaluating CMA-ES population with z worker processesspawn) max_workers mp_context initializerinitargscB<\SP\V44#r)rpmapr)r>pools&rrun_cma..sd488DY[d;e6frz!phase2: worker pool unavailable (z(); falling back to sequential evaluation)r)i)cmaCMAEvolutionStrategyrrr8r}rqstoprFrmp get_context concurrentfuturesProcessPoolExecutorrOSErrorPermissionErrorrun_cma)r"theta0sigma0r&rBrnr9r)rTrCrHr;r rLexcr0rAr1rQsf&&&f&ff @@@@rr^r^ss ! !&#,gvtYPR"S UB!S\"G 2667FItUD 2y' 3D  R R!| ?@ %C 3 5wwyy6se<= &  gy$/GEFF 4WI=NOP(J\    3 3!+Z 4  [)s$fg7799:3%@A *   K   K _ %\ /u4\ ]^wdCYZ[[\sB;0F +/E8E8.F 8 F F F GG:GGc D^RIHpVP;'g RV RV R2pV!RV 24\P!V4pVP ^,P 4pV!R4V!VP43/VBp V R,Pp V P\P4p \P!\V 4\PR7p ^p \\V 44F2pW,^8d!\V R , R 4W&V ^, p K0^p K4 WR &WR &\!V P#44p\%V P'44pV!R V R\V4 RVR RV^d,R R2 4\(R R .,#)aCompute in_long_position + bars_held_norm from the current winner policy. One-cycle circularity (documented in roadmap): the position state used for training comes from the OLD model's decisions, not the new model. This is acceptable for the first retrain; a second retrain with the new policy's state would converge. Mutates df in-place, adding columns 'in_long_position' and 'bars_held_norm'. Returns the extended feature_cols list. )generate_signalsz8results/winners/optimization_winner_strategy_mlp_scores_rz.csvz'approach-b: loading winner params from zNapproach-b: running generate_signals with old policy to get position state ...positionrrrin_long_positionbars_held_normu+approach-b: position features injected — r4z& bars in-position, max bars_held_norm=z.3fz (=.0fz bars))strategies.strategy_mlp_scoresrcapproach_b_paramsr#r$ilocto_dictrr:r;r8r<zerosrDrqrrrxrrr)r1assettfargsr9 _gen_signals params_path params_dfrsignalsrdin_long bars_heldbars_inrn_in_posmax_helds&&&&& r_inject_position_featuresrysP  W W EeWAbTQU V 1+?@ K(I ^^A  & & (FXY2779//Gz"))Hoobjj)GXbjj9IG 3x= ! ;?w4IL qLGG "%$7;;=!HY]]_%H 5hZqR J&s^3x|C.@ HI -/?@ @@rc\P!RR7pVPRRR7VPR\R^^.R7VPR\^ R 7VPR \R R 7VPR \R R 7VPR \^ R 7VPR\RR 7VPR\RR 7VPR.RORRR7VPR\^RR7VPR\\ RR7VPR\^*R 7VPRRRR 7VPR!RR"7VPR#R$R%7VPR&R'R%7VPR(R)R%7VPR*RR+R 7VPR,R-R%7VPR..ROR2R3R7VP 4p\P!4pR4pVP'd+VP'dVPVPrTM\VP4wrE\VP4pVP'EdVPR54PRR67p^^.Vn\#VP$^4Vn\#VP&^4Vn\#VP(^4VnVR7,P*\\-V4R8,4,pR9\/VP144\/VR7,P*R,P1443.p^^rTp MT\2P4p\2P6p \2P8p \:P<!VR,^,4p \?VP@VP(VPB4Vn VR7,P*R,p V!R:V R;V R<\-V4 R=VR7,P*^,P14 R>VR7,P*R,P14 R?VP 2 4VPD'd\GWdWQV4p M\Hp \KWm4p\MWaPN4wppR/^R0RR1^/pVPPR28wdVVPP,pR@VPR9dVR@,PTV8Hp\VPW44pVV,p\VPW44\Y^V4, ^d,pV!RAVPP RBV RC\VPW44 RDVRE RF2 4MV!RGVPP RH24\[WVVR7,VP VP$VP\WV4 wppVRI,pRJ/VCRKVPNRL\^RM\`RN\/V 4\/V 4./CROVPRP\bPd!\gVPRQ4Pi44Pk4RR\/VR7,P*^,P144\/VR7,P*R,P144.RSVP\RTVPPRU\lPn!\pPr4PuRVRW7/p\wVV4p\xPz!\xP|!V^<4\xP|!V^K4\xP|!V^4RX.4pV!RYVP#4RZ R[VPY4RZ R\V^,RZ R]V^,RZ R^V^,RZ 2 4VP~'Eg\V4p\xP!VV\, .4p\WnVVVWVPVP4 pV!V4wpp p!T!R_VR` RaV U"u.uFp"\V"^4NK up" 24\VVVPVP(VP&VP\W1P@4p#\-V4p$\V#Rb,R9V$V4pV#Rc,pRdVP(ReVPRfVP&RgVPRhV#PRiVP@4RjVPRkV#Rl,RmV#Rn,Ro\ \xP!V#Rn,44Rp\ \xP"!V#Rn,44/ VRq&VPR8XdV#PRr/4VRq,Rs&Rt\ V^,4Ru\ V^,4Rv\ V^,4Rw\ V^,4RxRyRzR{/VR|&VP;'g*\PPR}R~RRV RV R24p%\V%VWEVV R7V!RV% R\P!4V, RE R24V!RVR|, 24R9#uup"i)zTwo-phase MLP trainer) descriptionz--dataT)requiredz--hidden+)typenargsdefaultz --target-k)r~rz--phase1-epochsi,z--es-generationsz --es-popsizez --es-sigmarz--l2gMbP?z--fold-objectiverrzPhase-2 fold aggregation: mean matches legacy behavior; min emphasizes worst fold; mean_min blends 70%% mean / 30%% min; robust subtracts drawdown, P&L/DD, trade-count, negative-fold, and threshold-fragility penalties.)choicesrhelpz --es-workerszWorker processes for parallel CMA-ES population evaluation. Use 0 for auto, currently 50%% of logical CPUs capped by popsize; use 1 for sequential evaluation.)r~rrz--worker-fractionz.Logical CPU fraction used when --es-workers=0.z--seedz--smoke store_truez/Tiny run on recent bars for plumbing validation)actionrz --skip-phase2)rz--outzArtifact path override)rz--assetzHAsset override (e.g. COINBASE_BTCUSD) when filename format doesn't matchz--tfz?Timeframe override (e.g. 4H) when filename format doesn't matchz --approach-bzProbe: inject in_long_position + bars_held_norm (cols 50-51) computed from the current winner artifact's position state (one-cycle circularity per roadmap). Output artifact will have 52 features.z--approach-b-paramszMWinner CSV to derive position state from (default: auto-detect from asset/tf)z--regimebullbearsidewaysallzFilter Phase-1 supervised training to MVRV regime bars only (requires mvrv_regime column). Phase-2 CMA-ES always uses all IS bars.cT\R\P!R4 RV 2RR7R#)[z%H:%M:%Sz] T)flushN)printrstrftime)msgs&rr9main..log!s# $-- +,Bse4DAri r!rg?Nzasset=z tf=z bars=z window=rz hidden= mvrv_regimezregime=z: Phase-1 training bars u → z (rgz%)zWARNING: --regime=z0 requested but mvrv_regime not in data; ignoringrephase1target_k vol_window target_scaleval_span data_file data_sha256rb train_windowrn regime_filtercreatedseconds)timespecrzphase2: score range [z.1fz, z], seed thresholds entry=z exit=z conf=z"phase2: pretrain baseline fitness=r5r6rr( es_popsizees_sigmaes_generations l2_anchor es_workersr)r best_fitnessr! fold_calmarsrfold_mean_calmarfold_min_calmarphase2rrobust_diagnosticsrrrrrrrFrecommended_thresholds strategiesrmlp mlp_weights_rz.json)rm timeframetraining feature_colszartifact written: z (zs total)zrecommended thresholds: )rrrr)rrrr)QargparseArgumentParser add_argumentrrr parse_argsrrmrnrdatar3smoketailr0rr phase1_epochsrrrjrDr&dater, WFO_FOLDSWFO_MIN_OOS_TRADESWFO_MIN_VALID_FOLDSr# Timestampr rworker_fraction approach_bryrr rQrregimer%r:rxrrrnr@rGhashlibsha256openread hexdigestrnowrr isoformatr r8array percentile skip_phase2rrr r$rrr:r^rrrrrGrrjoinr )&prot0r9rmrnr1t75rrrrrrr^rOrP _REGIME_INT regime_int regime_maskbeforepctrp1_metare training_metascores0rrr_r"f0calmars0rr@r0rrGs& rmainrsD ,CDANN8dN+NN:CsRGNDNNRN8NN NN>Q;<NN&U,OPNN(giNN:'JTYab <<>D BB zzzdgggJJr#DII.  499 B zzz WWS\ % %4 % 0!f  !3!3R8!$"5"5q9dooq1jooc#b'D.12CHHJRZ__R-@-E-E-G)HIJ+,a    33 44LLr1. *4??DOOTMaMabDOjoob!G&tB4vc"gYh f:??1  " " $ %R6 (;(@(@(B'C8DKK= Z[ 0BcJ # "+AB .HAu1fb*a8K {{e - BJJ &]+22j@K%FK'Eeiik"SF^3c9C '$++&>vheCPUPYPYP[L\K]]_`cdg_hhjk l $T[[M1ab cqUBvJ #11499iRUWOFG 6?D ?W?j$--^\IG =? TYYw~~d499d&;&@&@&BCMMORZ__Q/4467RZ__R=P=U=U=W9XY 8<< -777K M!V$G gr" gr" gr" J   c2"W[[]34GH!!+As 36*Q-9LM1 c" $%    !&):+G HIrdHe.#224"&/Ha 0C92:;(QU1a[(;< > ?w t**DIIsOOM(m!$w-"5t<%[ $//:t}} d11; $((9doo> d11 DK DO bggd9o&> ? uRVVDO%<= # h   ( *<@HH]TV__doc__rconcurrent.futuresrYrjsonmultiprocessingrWrrsysrrrrr8pandasr#rruabspathrdirname__file__r,rhrrrrr r %strategies.strategy_activation_scoresr tools.worker_utilsr r r@rGr rrrrrr3rQrrrrrrrrrrr$r^ryr__name__rrrs@  ' RWW__X-F MNO DL    $""/ $<9FW=&/%d&C?L 3"I`-AhgNT zFr