<\/p>\n
Bir modelin veri seti \u00fczerindeki performans\u0131n\u0131 de\u011ferlendirmek i\u00e7in modelin yapt\u0131\u011f\u0131 tahminlerin g\u00f6zlemlenen verilerle ne kadar iyi e\u015fle\u015fti\u011fini \u00f6l\u00e7memiz gerekir.<\/span><\/p>\n Bunu yapmak i\u00e7in yayg\u0131n olarak kullan\u0131lan bir y\u00f6ntem, a\u015fa\u011f\u0131daki yakla\u015f\u0131m\u0131 kullanan k-katl\u0131 \u00e7apraz do\u011frulama<\/a> olarak bilinir:<\/span><\/p>\n 1.<\/strong> Bir veri k\u00fcmesini rastgele olarak yakla\u015f\u0131k olarak e\u015fit b\u00fcy\u00fckl\u00fckte k<\/em> gruba veya “katlamaya” b\u00f6l\u00fcn.<\/span><\/p>\n 2.<\/strong> Tutucu set olarak k\u0131vr\u0131mlardan birini se\u00e7in. \u015eablonu kalan k-1 k\u0131vr\u0131mlar\u0131na g\u00f6re ayarlay\u0131n. Gerilmi\u015f kattaki g\u00f6zlemler \u00fczerinden MSE testini hesaplay\u0131n.<\/span><\/p>\n 3.<\/strong> Her seferinde farkl\u0131 bir k\u00fcmeyi d\u0131\u015flama k\u00fcmesi olarak kullanarak bu i\u015flemi k<\/em> kez tekrarlay\u0131n.<\/span><\/p>\n 4.<\/strong> Genel test MSE’sini k<\/em> test MSE’sinin ortalamas\u0131 olarak hesaplay\u0131n.<\/span><\/p>\n Bu \u00f6\u011fretici, Python’da belirli bir model i\u00e7in k-katl\u0131 \u00e7apraz do\u011frulaman\u0131n nas\u0131l ger\u00e7ekle\u015ftirilece\u011fine ili\u015fkin ad\u0131m ad\u0131m bir \u00f6rnek sa\u011flar.<\/span><\/p>\n \u00d6ncelikle bu \u00f6rnek i\u00e7in gereken fonksiyonlar\u0131 ve k\u00fct\u00fcphaneleri y\u00fckleyece\u011fiz:<\/span><\/p>\n Daha sonra, x1<\/sub> ve x2<\/sub> olmak \u00fczere iki tahmin de\u011fi\u015fkeni ve tek bir yan\u0131t de\u011fi\u015fkeni y i\u00e7eren bir pandas DataFrame olu\u015fturaca\u011f\u0131z.<\/span><\/p>\n Daha sonra veri setine \u00e7oklu do\u011frusal regresyon modeli<\/a> yerle\u015ftirece\u011fiz ve modelin performans\u0131n\u0131 de\u011ferlendirmek i\u00e7in LOOCV ger\u00e7ekle\u015ftirece\u011fiz.<\/span><\/p>\n Sonu\u00e7tan ortalama mutlak hatan\u0131n (MAE) 3,614<\/strong> oldu\u011funu g\u00f6rebiliriz. Yani, model tahmini ile ger\u00e7ekte g\u00f6zlemlenen veriler aras\u0131ndaki ortalama mutlak hata 3,614’t\u00fcr.<\/span><\/p>\n Genel olarak MAE ne kadar d\u00fc\u015f\u00fck olursa, model ger\u00e7ek g\u00f6zlemleri o kadar iyi tahmin edebilir.<\/span><\/p>\n Model performans\u0131n\u0131 de\u011ferlendirmek i\u00e7in yayg\u0131n olarak kullan\u0131lan bir di\u011fer \u00f6l\u00e7\u00fcm, ortalama karek\u00f6k hatad\u0131r (RMSE). A\u015fa\u011f\u0131daki kod, bu metri\u011fin LOOCV kullan\u0131larak nas\u0131l hesaplanaca\u011f\u0131n\u0131 g\u00f6sterir:<\/span><\/p>\n Sonu\u00e7tan, ortalama kare hatas\u0131n\u0131n (RMSE) 4,284<\/strong> oldu\u011funu g\u00f6rebiliriz.<\/span><\/p>\n RMSE ne kadar d\u00fc\u015f\u00fck olursa, model ger\u00e7ek g\u00f6zlemleri o kadar iyi tahmin edebilir.<\/span><\/p>\n Uygulamada, genellikle birka\u00e7 farkl\u0131 model yerle\u015ftiririz ve hangi modelin en d\u00fc\u015f\u00fck test hatas\u0131 oranlar\u0131n\u0131 \u00fcretti\u011fine ve dolay\u0131s\u0131yla kullan\u0131lacak en iyi model oldu\u011funa karar vermek i\u00e7in her modelin RMSE veya MAE’sini kar\u015f\u0131la\u015ft\u0131r\u0131r\u0131z.<\/span><\/p>\n Ayr\u0131ca bu \u00f6rnekte k=5 katlamay\u0131 se\u00e7ti\u011fimizi, ancak istedi\u011finiz say\u0131da katlamay\u0131 se\u00e7ebilece\u011finizi unutmay\u0131n.<\/span><\/p>\n Pratikte genellikle 5 ile 10 kat aras\u0131nda se\u00e7im yapar\u0131z, \u00e7\u00fcnk\u00fc bu, g\u00fcvenilir test hatas\u0131 oranlar\u0131 \u00fcreten en uygun kat say\u0131s\u0131d\u0131r.<\/span><\/p>\n Sklearn’in KFold() i\u015flevine ili\u015fkin t\u00fcm belgeleri burada<\/a> bulabilirsiniz.<\/em><\/span><\/p>\n K-Fold \u00c7apraz Do\u011frulamaya Giri\u015f<\/a> Bir modelin veri seti \u00fczerindeki performans\u0131n\u0131 de\u011ferlendirmek i\u00e7in modelin yapt\u0131\u011f\u0131 tahminlerin g\u00f6zlemlenen verilerle ne kadar iyi e\u015fle\u015fti\u011fini \u00f6l\u00e7memiz gerekir. Bunu yapmak i\u00e7in yayg\u0131n olarak kullan\u0131lan bir y\u00f6ntem, a\u015fa\u011f\u0131daki yakla\u015f\u0131m\u0131 kullanan k-katl\u0131 \u00e7apraz do\u011frulama olarak bilinir: 1. Bir veri k\u00fcmesini rastgele olarak yakla\u015f\u0131k olarak e\u015fit b\u00fcy\u00fckl\u00fckte k gruba veya “katlamaya” b\u00f6l\u00fcn. 2. Tutucu set olarak k\u0131vr\u0131mlardan […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[11],"tags":[],"class_list":["post-1178","post","type-post","status-publish","format-standard","hentry","category-rehber"],"yoast_head":"\n Ad\u0131m 1: Gerekli Kitapl\u0131klar\u0131 Y\u00fckleyin<\/strong><\/span><\/h3>\n
from<\/span> sklearn. model_selection<\/span> import<\/span> train_test_split\nfrom<\/span> sklearn. model_selection<\/span> import<\/span> KFold\nfrom<\/span> sklearn. model_selection<\/span> import<\/span> cross_val_score\nfrom<\/span> sklearn. linear_model<\/span> import<\/span> LinearRegression\nfrom<\/span> numpy import<\/span> means\nfrom<\/span> numpy import<\/span> absolute\nfrom<\/span> numpy import<\/span> sqrt\nimport<\/span> pandas as<\/span> pd\n<\/strong><\/span><\/pre>\n 2. Ad\u0131m: Verileri olu\u015fturun<\/strong><\/span><\/h3>\n
df = pd.DataFrame({' y<\/span> ': [6, 8, 12, 14, 14, 15, 17, 22, 24, 23],\n ' x1<\/span> ': [2, 5, 4, 3, 4, 6, 7, 5, 8, 9],\n ' x2<\/span> ': [14, 12, 12, 13, 7, 8, 7, 4, 6, 5]})\n<\/strong><\/span><\/pre>\n Ad\u0131m 3: K Katlamal\u0131 \u00c7apraz Do\u011frulama Ger\u00e7ekle\u015ftirin<\/strong><\/span><\/h3>\n
#define predictor and response variables\n<\/span>X = df[[' x1<\/span> ', ' x2<\/span> ']]\ny = df[' y<\/span> ']\n\n#define cross-validation method to use\n<\/span>cv<\/span> =<\/span> KFold<\/span> (<\/span> n_splits<\/span> =<\/span> 10<\/span> ,<\/span> random_state<\/span> =<\/span> 1<\/span> ,<\/span> shuffle<\/span> =<\/span> True<\/span> )<\/span>\n\n#build multiple linear regression model\n<\/span>model = LinearRegression()\n\n#use k-fold CV to evaluate model\n<\/span>scores = cross_val_score(model, X, y, scoring=' neg_mean_absolute_error<\/span> ',\n cv=cv, n_jobs=-1)\n\n#view mean absolute error\n<\/span>mean(absolute(scores))\n\n3.6141267491803646\n<\/strong><\/span><\/pre>\n #define predictor and response variables\n<\/span>X = df[[' x1<\/span> ', ' x2<\/span> ']]\ny = df[' y<\/span> ']\n\n#define cross-validation method to use\n<\/span>cv<\/span> =<\/span> KFold<\/span> (<\/span> n_splits<\/span> =<\/span> 5<\/span> ,<\/span> random_state<\/span> =<\/span> 1<\/span> ,<\/span> shuffle<\/span> =<\/span> True<\/span> )<\/span> \n\n#build multiple linear regression model\n<\/span>model = LinearRegression()\n\n#use LOOCV to evaluate model\n<\/span>scores = cross_val_score(model, X, y, scoring=' neg_mean_squared_error<\/span> ',\n cv=cv, n_jobs=-1)\n\n#view RMSE\n<\/span>sqrt(mean(absolute(scores)))\n\n4.284373111711816<\/strong><\/span><\/pre>\n Ek kaynaklar<\/strong><\/span><\/h3>\n
Python’da Do\u011frusal Regresyon \u0130\u00e7in Tam Bir K\u0131lavuz<\/a>
Python’da Birini D\u0131\u015far\u0131da B\u0131rakma \u00c7apraz Do\u011frulama<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"