[{"data":1,"prerenderedAt":1263},["ShallowReactive",2],{"lang-switch-post-\u002Fplaylists\u002Fmachine-learning-specialization\u002Fw3-lab05-cost-function":3,"post-pt-machine-learning-specialization-w3-lab05-cost-function":4},"\u002Fen\u002Fplaylists\u002Fmachine-learning-specialization\u002Fw3-lab05-cost-function",{"id":5,"title":6,"body":7,"cover":1246,"date":1247,"description":1248,"extension":1249,"meta":1250,"navigation":1251,"order":1252,"path":1253,"playlist":1254,"seo":1255,"status":1256,"stem":1257,"tags":1258,"__hash__":1262},"posts\u002Fpt\u002Fplaylists\u002Fmachine-learning-specialization\u002Fw3-lab05-cost-function.md","Da Perda pro Custo: Somando o Erro de Todo Mundo",{"type":8,"value":9,"toc":1239},"minimark",[10,40,45,886,897,901,965,1022,1026,1041,1056,1072,1075,1086,1101,1105,1153,1157,1185,1200,1212,1235],[11,12,13,18,19,23,24,27,28,32,33,27,36,39],"p",{},[14,15,17],"a",{"href":16},"\u002Fplaylists\u002Fmachine-learning-specialization\u002Fw3-lab04-logistic-loss","A perda logística"," mede o erro de ",[20,21,22],"strong",{},"um"," exemplo. O curso já define essa distinção: ",[20,25,26],{},"perda"," (",[29,30,31],"em",{},"loss",") é sobre um exemplo só, ",[20,34,35],{},"custo",[29,37,38],{},"cost",") é a perda somada (com a média) sobre o conjunto de treino inteiro. É o custo que gradiente descendente minimiza.",[41,42,44],"h2",{"id":43},"a-função-de-custo-completa","A função de custo completa",[11,46,47],{},[48,49,52,263],"span",{"className":50},[51],"katex",[48,53,56],{"className":54},[55],"katex-mathml",[57,58,60],"math",{"xmlns":59},"http:\u002F\u002Fwww.w3.org\u002F1998\u002FMath\u002FMathML",[61,62,63,258],"semantics",{},[64,65,66,70,75,79,83,86,89,92,102,127],"mrow",{},[67,68,69],"mi",{},"J",[71,72,74],"mo",{"stretchy":73},"false","(",[67,76,78],{"mathvariant":77},"bold","w",[71,80,82],{"separator":81},"true",",",[67,84,85],{},"b",[71,87,88],{"stretchy":73},")",[71,90,91],{},"=",[93,94,95,99],"mfrac",{},[96,97,98],"mn",{},"1",[67,100,101],{},"m",[103,104,105,108,118],"msubsup",{},[71,106,107],{},"∑",[64,109,110,113,115],{},[67,111,112],{},"i",[71,114,91],{},[96,116,117],{},"0",[64,119,120,122,125],{},[67,121,101],{},[71,123,124],{},"−",[96,126,98],{},[64,128,129,132,134,148,151,154,191,193,195,197,199,211,213,215,217,255],{},[71,130,131],{"fence":81},"[",[71,133,124],{},[135,136,137,140],"msup",{},[67,138,139],{},"y",[64,141,142,144,146],{},[71,143,74],{"stretchy":73},[67,145,112],{},[71,147,88],{"stretchy":73},[67,149,150],{},"log",[71,152,153],{},"⁡",[64,155,156,158,172,174,187,189],{},[71,157,74],{"fence":81},[159,160,161,164],"msub",{},[67,162,163],{},"f",[64,165,166,168,170],{},[67,167,78],{"mathvariant":77},[71,169,82],{"separator":81},[67,171,85],{},[71,173,74],{"stretchy":73},[135,175,176,179],{},[67,177,178],{"mathvariant":77},"x",[64,180,181,183,185],{},[71,182,74],{"stretchy":73},[67,184,112],{},[71,186,88],{"stretchy":73},[71,188,88],{"stretchy":73},[71,190,88],{"fence":81},[71,192,124],{},[71,194,74],{"stretchy":73},[96,196,98],{},[71,198,124],{},[135,200,201,203],{},[67,202,139],{},[64,204,205,207,209],{},[71,206,74],{"stretchy":73},[67,208,112],{},[71,210,88],{"stretchy":73},[71,212,88],{"stretchy":73},[67,214,150],{},[71,216,153],{},[64,218,219,221,223,225,237,239,251,253],{},[71,220,74],{"fence":81},[96,222,98],{},[71,224,124],{},[159,226,227,229],{},[67,228,163],{},[64,230,231,233,235],{},[67,232,78],{"mathvariant":77},[71,234,82],{"separator":81},[67,236,85],{},[71,238,74],{"stretchy":73},[135,240,241,243],{},[67,242,178],{"mathvariant":77},[64,244,245,247,249],{},[71,246,74],{"stretchy":73},[67,248,112],{},[71,250,88],{"stretchy":73},[71,252,88],{"stretchy":73},[71,254,88],{"fence":81},[71,256,257],{"fence":81},"]",[259,260,262],"annotation",{"encoding":261},"application\u002Fx-tex","J(\\mathbf{w},b) = \\frac{1}{m}\\sum_{i=0}^{m-1}\\left[-y^{(i)}\\log\\left(f_{\\mathbf{w},b}(\\mathbf{x}^{(i)})\\right) - (1-y^{(i)})\\log\\left(1-f_{\\mathbf{w},b}(\\mathbf{x}^{(i)})\\right)\\right]",[48,264,267,318],{"className":265,"ariaHidden":81},[266],"katex-html",[48,268,271,276,282,286,291,295,300,303,307,311,315],{"className":269},[270],"base",[48,272],{"className":273,"style":275},[274],"strut","height:1em;vertical-align:-0.25em;",[48,277,69],{"className":278,"style":281},[279,280],"mord","mathnormal","margin-right:0.0962em;",[48,283,74],{"className":284},[285],"mopen",[48,287,78],{"className":288,"style":290},[279,289],"mathbf","margin-right:0.016em;",[48,292,82],{"className":293},[294],"mpunct",[48,296],{"className":297,"style":299},[298],"mspace","margin-right:0.1667em;",[48,301,85],{"className":302},[279,280],[48,304,88],{"className":305},[306],"mclose",[48,308],{"className":309,"style":310},[298],"margin-right:0.2778em;",[48,312,91],{"className":313},[314],"mrel",[48,316],{"className":317,"style":310},[298],[48,319,321,325,413,416,496,499],{"className":320},[270],[48,322],{"className":323,"style":324},[274],"height:1.304em;vertical-align:-0.35em;",[48,326,328,332,410],{"className":327},[279],[48,329],{"className":330},[285,331],"nulldelimiter",[48,333,335],{"className":334},[93],[48,336,340,401],{"className":337},[338,339],"vlist-t","vlist-t2",[48,341,344,396],{"className":342},[343],"vlist-r",[48,345,349,370,381],{"className":346,"style":348},[347],"vlist","height:0.8451em;",[48,350,352,357],{"style":351},"top:-2.655em;",[48,353],{"className":354,"style":356},[355],"pstrut","height:3em;",[48,358,364],{"className":359},[360,361,362,363],"sizing","reset-size6","size3","mtight",[48,365,367],{"className":366},[279,363],[48,368,101],{"className":369},[279,280,363],[48,371,373,376],{"style":372},"top:-3.23em;",[48,374],{"className":375,"style":356},[355],[48,377],{"className":378,"style":380},[379],"frac-line","border-bottom-width:0.04em;",[48,382,384,387],{"style":383},"top:-3.394em;",[48,385],{"className":386,"style":356},[355],[48,388,390],{"className":389},[360,361,362,363],[48,391,393],{"className":392},[279,363],[48,394,98],{"className":395},[279,363],[48,397,400],{"className":398},[399],"vlist-s","​",[48,402,404],{"className":403},[343],[48,405,408],{"className":406,"style":407},[347],"height:0.345em;",[48,409],{},[48,411],{"className":412},[306,331],[48,414],{"className":415,"style":299},[298],[48,417,420,426],{"className":418},[419],"mop",[48,421,107],{"className":422,"style":425},[419,423,424],"op-symbol","small-op","position:relative;top:0em;",[48,427,430],{"className":428},[429],"msupsub",[48,431,433,487],{"className":432},[338,339],[48,434,436,484],{"className":435},[343],[48,437,440,462],{"className":438,"style":439},[347],"height:0.954em;",[48,441,443,447],{"style":442},"top:-2.4003em;margin-left:0em;margin-right:0.05em;",[48,444],{"className":445,"style":446},[355],"height:2.7em;",[48,448,450],{"className":449},[360,361,362,363],[48,451,453,456,459],{"className":452},[279,363],[48,454,112],{"className":455},[279,280,363],[48,457,91],{"className":458},[314,363],[48,460,117],{"className":461},[279,363],[48,463,465,468],{"style":464},"top:-3.2029em;margin-right:0.05em;",[48,466],{"className":467,"style":446},[355],[48,469,471],{"className":470},[360,361,362,363],[48,472,474,477,481],{"className":473},[279,363],[48,475,101],{"className":476},[279,280,363],[48,478,124],{"className":479},[480,363],"mbin",[48,482,98],{"className":483},[279,363],[48,485,400],{"className":486},[399],[48,488,490],{"className":489},[343],[48,491,494],{"className":492,"style":493},[347],"height:0.2997em;",[48,495],{},[48,497],{"className":498,"style":299},[298],[48,500,503,513,516,557,560,568,571,683,687,690,693,696,699,702,705,708,746,749,752,757,760,880],{"className":501},[502],"minner",[48,504,508],{"className":505,"style":507},[285,506],"delimcenter","top:0em;",[48,509,131],{"className":510},[511,512],"delimsizing","size1",[48,514,124],{"className":515},[279],[48,517,519,523],{"className":518},[279],[48,520,139],{"className":521,"style":522},[279,280],"margin-right:0.0359em;",[48,524,526],{"className":525},[429],[48,527,529],{"className":528},[338],[48,530,532],{"className":531},[343],[48,533,536],{"className":534,"style":535},[347],"height:0.888em;",[48,537,539,542],{"style":538},"top:-3.063em;margin-right:0.05em;",[48,540],{"className":541,"style":446},[355],[48,543,545],{"className":544},[360,361,362,363],[48,546,548,551,554],{"className":547},[279,363],[48,549,74],{"className":550},[285,363],[48,552,112],{"className":553},[279,280,363],[48,555,88],{"className":556},[306,363],[48,558],{"className":559,"style":299},[298],[48,561,563,564],{"className":562},[419],"lo",[48,565,567],{"style":566},"margin-right:0.0139em;","g",[48,569],{"className":570,"style":299},[298],[48,572,574,580,633,636,674,677],{"className":573},[502],[48,575,577],{"className":576,"style":507},[285,506],[48,578,74],{"className":579},[511,512],[48,581,583,587],{"className":582},[279],[48,584,163],{"className":585,"style":586},[279,280],"margin-right:0.1076em;",[48,588,590],{"className":589},[429],[48,591,593,624],{"className":592},[338,339],[48,594,596,621],{"className":595},[343],[48,597,600],{"className":598,"style":599},[347],"height:0.3361em;",[48,601,603,606],{"style":602},"top:-2.55em;margin-left:-0.1076em;margin-right:0.05em;",[48,604],{"className":605,"style":446},[355],[48,607,609],{"className":608},[360,361,362,363],[48,610,612,615,618],{"className":611},[279,363],[48,613,78],{"className":614,"style":290},[279,289,363],[48,616,82],{"className":617},[294,363],[48,619,85],{"className":620},[279,280,363],[48,622,400],{"className":623},[399],[48,625,627],{"className":626},[343],[48,628,631],{"className":629,"style":630},[347],"height:0.2861em;",[48,632],{},[48,634,74],{"className":635},[285],[48,637,639,642],{"className":638},[279],[48,640,178],{"className":641},[279,289],[48,643,645],{"className":644},[429],[48,646,648],{"className":647},[338],[48,649,651],{"className":650},[343],[48,652,654],{"className":653,"style":535},[347],[48,655,656,659],{"style":538},[48,657],{"className":658,"style":446},[355],[48,660,662],{"className":661},[360,361,362,363],[48,663,665,668,671],{"className":664},[279,363],[48,666,74],{"className":667},[285,363],[48,669,112],{"className":670},[279,280,363],[48,672,88],{"className":673},[306,363],[48,675,88],{"className":676},[306],[48,678,680],{"className":679,"style":507},[306,506],[48,681,88],{"className":682},[511,512],[48,684],{"className":685,"style":686},[298],"margin-right:0.2222em;",[48,688,124],{"className":689},[480],[48,691],{"className":692,"style":686},[298],[48,694,74],{"className":695},[285],[48,697,98],{"className":698},[279],[48,700],{"className":701,"style":686},[298],[48,703,124],{"className":704},[480],[48,706],{"className":707,"style":686},[298],[48,709,711,714],{"className":710},[279],[48,712,139],{"className":713,"style":522},[279,280],[48,715,717],{"className":716},[429],[48,718,720],{"className":719},[338],[48,721,723],{"className":722},[343],[48,724,726],{"className":725,"style":535},[347],[48,727,728,731],{"style":538},[48,729],{"className":730,"style":446},[355],[48,732,734],{"className":733},[360,361,362,363],[48,735,737,740,743],{"className":736},[279,363],[48,738,74],{"className":739},[285,363],[48,741,112],{"className":742},[279,280,363],[48,744,88],{"className":745},[306,363],[48,747,88],{"className":748},[306],[48,750],{"className":751,"style":299},[298],[48,753,563,755],{"className":754},[419],[48,756,567],{"style":566},[48,758],{"className":759,"style":299},[298],[48,761,763,769,772,775,778,781,830,833,871,874],{"className":762},[502],[48,764,766],{"className":765,"style":507},[285,506],[48,767,74],{"className":768},[511,512],[48,770,98],{"className":771},[279],[48,773],{"className":774,"style":686},[298],[48,776,124],{"className":777},[480],[48,779],{"className":780,"style":686},[298],[48,782,784,787],{"className":783},[279],[48,785,163],{"className":786,"style":586},[279,280],[48,788,790],{"className":789},[429],[48,791,793,822],{"className":792},[338,339],[48,794,796,819],{"className":795},[343],[48,797,799],{"className":798,"style":599},[347],[48,800,801,804],{"style":602},[48,802],{"className":803,"style":446},[355],[48,805,807],{"className":806},[360,361,362,363],[48,808,810,813,816],{"className":809},[279,363],[48,811,78],{"className":812,"style":290},[279,289,363],[48,814,82],{"className":815},[294,363],[48,817,85],{"className":818},[279,280,363],[48,820,400],{"className":821},[399],[48,823,825],{"className":824},[343],[48,826,828],{"className":827,"style":630},[347],[48,829],{},[48,831,74],{"className":832},[285],[48,834,836,839],{"className":835},[279],[48,837,178],{"className":838},[279,289],[48,840,842],{"className":841},[429],[48,843,845],{"className":844},[338],[48,846,848],{"className":847},[343],[48,849,851],{"className":850,"style":535},[347],[48,852,853,856],{"style":538},[48,854],{"className":855,"style":446},[355],[48,857,859],{"className":858},[360,361,362,363],[48,860,862,865,868],{"className":861},[279,363],[48,863,74],{"className":864},[285,363],[48,866,112],{"className":867},[279,280,363],[48,869,88],{"className":870},[306,363],[48,872,88],{"className":873},[306],[48,875,877],{"className":876,"style":507},[306,506],[48,878,88],{"className":879},[511,512],[48,881,883],{"className":882,"style":507},[306,506],[48,884,257],{"className":885},[511,512],[11,887,888,889,892,893,896],{},"Parece complicada, mas é só ",[14,890,891],{"href":16},"a perda de um exemplo",", a mesma fórmula de linha única do lab passado, dentro de um somatório dividido por ",[894,895,101],"code",{},". Média da perda de todo mundo.",[41,898,900],{"id":899},"o-código-linha-a-linha","O código, linha a linha",[902,903,908],"pre",{"className":904,"code":905,"language":906,"meta":907,"style":907},"language-python shiki shiki-themes github-light github-dark","def compute_cost_logistic(X, y, w, b):\n    m = X.shape[0]\n    cost = 0.0\n    for i in range(m):\n        z_i = np.dot(X[i],w) + b\n        f_wb_i = sigmoid(z_i)\n        cost +=  -y[i]*np.log(f_wb_i) - (1-y[i])*np.log(1-f_wb_i)\n    cost = cost \u002F m\n    return cost\n","python","",[894,909,910,917,923,929,935,941,947,953,959],{"__ignoreMap":907},[48,911,914],{"class":912,"line":913},"line",1,[48,915,916],{},"def compute_cost_logistic(X, y, w, b):\n",[48,918,920],{"class":912,"line":919},2,[48,921,922],{},"    m = X.shape[0]\n",[48,924,926],{"class":912,"line":925},3,[48,927,928],{},"    cost = 0.0\n",[48,930,932],{"class":912,"line":931},4,[48,933,934],{},"    for i in range(m):\n",[48,936,938],{"class":912,"line":937},5,[48,939,940],{},"        z_i = np.dot(X[i],w) + b\n",[48,942,944],{"class":912,"line":943},6,[48,945,946],{},"        f_wb_i = sigmoid(z_i)\n",[48,948,950],{"class":912,"line":949},7,[48,951,952],{},"        cost +=  -y[i]*np.log(f_wb_i) - (1-y[i])*np.log(1-f_wb_i)\n",[48,954,956],{"class":912,"line":955},8,[48,957,958],{},"    cost = cost \u002F m\n",[48,960,962],{"class":912,"line":961},9,[48,963,964],{},"    return cost\n",[966,967,968,975,985,1000,1006,1015],"ul",{},[969,970,971,974],"li",{},[894,972,973],{},"m = X.shape[0]",": quantos exemplos tem no conjunto.",[969,976,977,978,981,982,984],{},"o ",[894,979,980],{},"for"," percorre cada paciente ",[894,983,112],{},".",[969,986,987,990,991,995,996,999],{},[894,988,989],{},"z_i = np.dot(X[i],w) + b",": a parte linear, ",[14,992,994],{"href":993},"\u002Fplaylists\u002Fmachine-learning-specialization\u002Fw3-lab03-decision-boundary","a mesma combinação de sempre",", só que agora ",[894,997,998],{},"X[i]"," pode ter várias variáveis, então é produto escalar em vez de multiplicação simples.",[969,1001,1002,1005],{},[894,1003,1004],{},"f_wb_i = sigmoid(z_i)",": espreme pra virar probabilidade.",[969,1007,1008,1009,1011,1012,984],{},"a linha do meio soma a perda desse paciente ao acumulador ",[894,1010,38],{},", ",[14,1013,1014],{"href":16},"a mesma fórmula de linha única do lab passado",[969,1016,1017,1018,1021],{},"no fim, ",[894,1019,1020],{},"cost \u002F m"," faz a média.",[41,1023,1025],{"id":1024},"testando-com-dois-valores-de-b-diferentes","Testando com dois valores de b diferentes",[11,1027,1028,1029,1032,1033,1036,1037,1040],{},"Mesmo dataset de sempre, duas fronteiras candidatas: ",[894,1030,1031],{},"b=-3"," e ",[894,1034,1035],{},"b=-4"," (mantendo ",[894,1038,1039],{},"w=[1,1]"," fixo nos dois).",[902,1042,1044],{"className":904,"code":1043,"language":906,"meta":907,"style":907},"print(\"Custo com b = -3:\", compute_cost_logistic(X_train, y_train, np.array([1,1]), -3))\nprint(\"Custo com b = -4:\", compute_cost_logistic(X_train, y_train, np.array([1,1]), -4))\n",[894,1045,1046,1051],{"__ignoreMap":907},[48,1047,1048],{"class":912,"line":913},[48,1049,1050],{},"print(\"Custo com b = -3:\", compute_cost_logistic(X_train, y_train, np.array([1,1]), -3))\n",[48,1052,1053],{"class":912,"line":919},[48,1054,1055],{},"print(\"Custo com b = -4:\", compute_cost_logistic(X_train, y_train, np.array([1,1]), -4))\n",[1057,1058,1059],"blockquote",{},[11,1060,1061,1064,1065,1068,1069],{},[20,1062,1063],{},"Saída:"," ",[894,1066,1067],{},"Custo com b = -3: 0.3669"," · ",[894,1070,1071],{},"Custo com b = -4: 0.5037",[11,1073,1074],{},"O custo já denuncia qual fronteira é pior, antes de eu precisar olhar gráfico nenhum. Mas o gráfico confirma:",[1076,1077],"classification-boundary-comparison-chart",{":boundaries":1078,":classes":1079,":points":1080,":x-max":1081,":x-min":117,":y-max":1081,":y-min":117,"negative-label":1082,"positive-label":1083,"x-label":1084,"y-label":1085},"[{\"w\":[1,1],\"bias\":-3,\"label\":\"b=-3 (custo 0.3669)\"},{\"w\":[1,1],\"bias\":-4,\"label\":\"b=-4 (custo 0.5037)\",\"dashed\":true}]","[-1,-1,-1,1,1,1]","[[0.5,1.5],[1,1],[1.5,0.5],[3,0.5],[2,2],[1,2.5]]","4","y=0","y=1","x0","x1",[11,1087,1088,1089,1091,1092,1094,1095,1097,1098,1100],{},"A linha de ",[894,1090,1035],{}," (tracejada) fica deslocada pra cima e pra direita em relação à de ",[894,1093,1031],{},", e nesse deslocamento ela passa a deixar dois pacientes ",[894,1096,1083],{}," do lado errado (abaixo da linha, previstos como ",[894,1099,117],{},"). Faz sentido o custo dela ser maior: custo alto é literalmente o modelo dizendo \"essa fronteira explica pior os dados que eu vi\".",[41,1102,1104],{"id":1103},"fechando","Fechando",[1106,1107,1108,1122],"table",{},[1109,1110,1111],"thead",{},[1112,1113,1114,1119],"tr",{},[1115,1116,1118],"th",{"align":1117},"left","O que eu já sabia",[1115,1120,1121],{"align":1117},"O que essa aula assentou",[1123,1124,1125,1134,1145],"tbody",{},[1112,1126,1127,1131],{},[1128,1129,1130],"td",{"align":1117},"Perda mede o erro de um exemplo",[1128,1132,1133],{"align":1117},"Custo é a média da perda sobre todo o conjunto de treino, é isso que se minimiza",[1112,1135,1136,1139],{},[1128,1137,1138],{"align":1117},"A fórmula da perda logística, de linha única",[1128,1140,1141,1142,1144],{"align":1117},"O mesmo código, dentro de um ",[894,1143,980],{}," com média no fim, vira a função de custo completa",[1112,1146,1147,1150],{},[1128,1148,1149],{"align":1117},"Uma fronteira pior classifica mais gente errado",[1128,1151,1152],{"align":1117},"O custo mede essa piora numericamente, antes mesmo de eu plotar qualquer gráfico",[41,1154,1156],{"id":1155},"aplicação-prática","Aplicação Prática",[11,1158,1159,1160,27,1163,1166,1167,1170,1171,1174,1175,1177,1178,1180,1181,1184],{},"Uso o modelo real de duas variáveis ",[14,1161,1162],{"href":993},"do lab da fronteira de decisão",[894,1164,1165],{},"worst concave points"," + ",[894,1168,1169],{},"worst perimeter",", ambas normalizadas), com os pesos ",[894,1172,1173],{},"w=[2.5494, 4.0739]"," treinados por gradiente descendente, e comparo o custo do ",[894,1176,85],{}," treinado contra um ",[894,1179,85],{}," deliberadamente pior (deslocado em ",[894,1182,1183],{},"+2","):",[902,1186,1188],{"className":904,"code":1187,"language":906,"meta":907,"style":907},"print(\"Custo com b treinado:\", compute_cost_logistic(X_train_z, y_train, w, b))\nprint(\"Custo com b + 2:\", compute_cost_logistic(X_train_z, y_train, w, b + 2))\n",[894,1189,1190,1195],{"__ignoreMap":907},[48,1191,1192],{"class":912,"line":913},[48,1193,1194],{},"print(\"Custo com b treinado:\", compute_cost_logistic(X_train_z, y_train, w, b))\n",[48,1196,1197],{"class":912,"line":919},[48,1198,1199],{},"print(\"Custo com b + 2:\", compute_cost_logistic(X_train_z, y_train, w, b + 2))\n",[1057,1201,1202],{},[11,1203,1204,1064,1206,1068,1209],{},[20,1205,1063],{},[894,1207,1208],{},"Custo com b treinado: 0.1461",[894,1210,1211],{},"Custo com b + 2: 0.2391",[11,1213,1214,1215,1217,1218,1221,1222,1224,1225,1228,1229,1231,1232,1234],{},"O custo mais que soma um terço a mais só por deslocar a fronteira em duas unidades. E a acurácia sente isso na prática: com o ",[894,1216,85],{}," treinado, acurácia de treino é ",[894,1219,1220],{},"0.9319","; com o ",[894,1223,85],{}," deslocado, cai pra ",[894,1226,1227],{},"0.8989",". Um ",[894,1230,85],{}," \"quase certo\" (a direção da fronteira, dada por ",[894,1233,78],{},", continua a mesma) já perde 3 pontos percentuais de acurácia, e o custo denuncia exatamente essa piora, do jeito que gradiente descendente enxerga: como um número maior, que ele vai tentar diminuir de novo nos próximos passos.",[1236,1237,1238],"style",{},"html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":907,"searchDepth":919,"depth":919,"links":1240},[1241,1242,1243,1244,1245],{"id":43,"depth":919,"text":44},{"id":899,"depth":919,"text":900},{"id":1024,"depth":919,"text":1025},{"id":1103,"depth":919,"text":1104},{"id":1155,"depth":919,"text":1156},null,"2026-09-04","Semana 3, Lab 5: a perda logística mede o erro de um paciente só. O custo soma essa perda pra todo o conjunto de treino numa única função, a que gradiente descendente vai minimizar. Um exemplo real mostra como o custo enxerga uma fronteira pior antes mesmo de eu olhar o gráfico.","md",{},true,16,"\u002Fpt\u002Fplaylists\u002Fmachine-learning-specialization\u002Fw3-lab05-cost-function","machine-learning-specialization",{"title":6,"description":1248},"published","pt\u002Fplaylists\u002Fmachine-learning-specialization\u002Fw3-lab05-cost-function",[1259,1260,1261],"classificacao","regressao-logistica","funcao-de-custo","9RDZZxksPGl9Mx3TD_Ru28RWKRgUo6_WB3croys97yE",1788965098015]