a j @s4dZddlZddlZddlZddlZddlZddlm Z ddl Z e dddl m ZejjdddedgZed D]*Zed r~ed kr~eed eq~eeZd dZdD]d?d@dAZdHZ?e9e9dIe?kZ@dJdKZAeAdLddMddNddOdeAdPddIddQddRdeAdSddMddTddUdgZBe:dVe:dWdXdYdZd[d\d]d^d_d`dad\dbd`dcdde:deeBD]~ZCeCdbdfkrdgndhZDe:eCdidXeCdjdZdkeCdldmdneCdodpeCdqdreCdadseCdbdteDeCdudv qe:dwdxdydzd{ZEejFd|d}\ZGZHeBD]RZCeHjIeJe;eCd~eCd~eEeCdideCdideCduddeCdjddqheHjKddddddeHLdeHMdeHNdeHjOddeHjPd8ddeGQeGjRdddeSeGe:ddS)aP Corrected re-run of research.ipynb (binned_df path). Faithfully reproduces the notebook pipeline (cells 3-24: load live_data, highest-CLV pick, time-bin target selection from first 20%, exclude USL/BSA, dedup one snapshot per game near TARGET_MINS) then applies ONLY the EVALUATION fixes (core strategy untouched: draws KEPT, original Kalshi-CLV floor >0.04, NO upper cap): 1. NET returns: win = decimal_odds - 1 (not the gross decimal odds). 2. Significance run on the NET returns (correct breakeven = 0). 3. Kalshi Kelly uses pin_fair of the bet side as p_win (not the market ask). N)statsZAggzutf-8replace)encodingerrorsz E:\soccerZ live_dataz.csvztest.csvz live_data/cCs>|||d}t|dr$dtjfSt||jd}|||fS)NHDAr)key)pdZisnanpnanmaxget)hdaZcdkr.E:\soccer\research\analysis\_research_fixed.py highest_clvs   r)b365kalshicCs4t|dtd|dtd|dtddS)Nclv__home_draw_awayrrbkrrrr&r!ZaxisZ _highest_clvcCs4t|dtd|dtd|dtddS)Nrrrrr#rrrrrr!(r"Z_highest_clv_valZftrb365_highest_clvZ correct_b365kalshi_highest_clvcorrect_kalshi)ZsubsetZstatusZPREelapsed)binscCsFtttdD]0}t||kr4t|dkrnq|SqdSNr#)rangelenr*)xirrr assign_bin1s$r0bing?meancountcCstt|t|ddS)Nr#)roundr*r.rrrr!:r"Zbin_mid)r'r2F)Z ascendingilocZleagueZUSLZBSAZdisthomeZawayfirst)Zkeep timestampcCst|S)N)r Z to_datetimer:r6rrrr!Br"T)Zdrop)Z result_homeZ result_awayZclv_kalshi_homeZclv_kalshi_drawZclv_kalshi_awayz'TARGET_MINS (best bin from first 20%): z#games after dedup/exclude USL+BSA:  rrrr皙?cCs0|d}||d||}tdt|||S)Nr#g)rmin)poddsbrmfbfrrr kelly_sizeKsrD@@c CsV|}g}t|||D]<\}}} t||||} || r>| |dn| 7}||q|Sr+)ziprDappend) r?ZprobsZoutcomesZbr0rAr@eqor>wZbetrrrcompoundPs  rKg{Gz?Zkalshi_highest_clv_valc Cst}|j|dd}|j|dd}||j} t| dk|dd} t|j|j| } | | jddt t | } dt j | } | }t|t ||| d| | | | d| d S)zLStrategy on its own pick side; original Kalshi-CLV gate (>0.04), draws kept.r#r$g)Zddofd) namenwrnet_sumroizr>finalrH)gatecopyapplyvaluesr whererKr2ZstdZsqrtr-rZnormZcdfdictsum)rNZpick_colZ clv_val_colZ correct_colZodds_fnZ pinfair_fnsr?ZpinZwonZnetrHrSr>rPrrrevaluate[s "r]B365Zb365_highest_clv_valcCs|dt|dS)Nrr%siderrrrr!mr"cCs|dt|dSNpin_fairr%r_rrrrr!nr"KalshicCsd|dt|ddS)N?rr&_askr_rrrrr!qr"cCs|dt|dS)Nrbr&r_rrrrr!rr"Kalshi w/ B365cCsd|dt|ddS)Nrdrr%rer_rrrrr!ur"cCs|dt|dSrar_rrrrr!vr"z`=== CORRECTED (net returns, pin_fair Kelly) | core unchanged: draws KEPT, floor>0.04, no cap ===Zstrategyz>16Nz>5zwin%z>7ZnetSumz>9zROI%z>8rSr>zKelly$z>10zG-----------------------------------------------------------------------g?* rNrOrLrPz>6.1f%rQz>+9.1frRz>+8.1fz>+7.2fz>8.4frTz>9.0fzR (Kelly$ = final compounded bankroll from $1000, real Kelly w/ pin_fair, 2.5% cap)z#2077b4z#d62728z#2ca02c)r^rcrf) g@)ZfigsizerHr4z ($z,.0fz, n=))colorlwZlabeliZgrayz--g333333?)rmZlsrnalphaZBetsz Bankroll ($)zGCorrected compounded returns (net, pin_fair Kelly) - draws kept, no capbest)Zlocg333333?)roztest_fixed.pngx)Zdpiz chart: test_fixed.png)r<)rEr<)T__doc__osZmathsysZpandasr Znumpyr ZscipyrZ matplotlibZuseZmatplotlib.pyplotZpyplotZpltstdout reconfigurechdirZdfslistdirrCendswithrGZread_csvconcatZdfrrrWZastypeintZdropnaZddZceilabsr=Znum_binsZ histogramr*r0r5shapeZfirst_20headZdf_firstgroupbyZaggZ reset_indexZbinnedZ sort_valuesr7rphasattrZ TARGET_MINSrVZdf_remZdrop_duplicatesgprintr-r`rDrKZFLOORrUr]Zresultsr ZsigZcolorsZsubplotsZfigZaxZplotr,ZaxhlineZ set_xlabelZ set_ylabelZ set_titleZlegendZgridZ tight_layoutZsavefigcloserrrrs        .   80  $"  "