{"id":664,"date":"2021-02-21T10:21:41","date_gmt":"2021-02-21T01:21:41","guid":{"rendered":"http:\/\/mukgee.com\/?p=664"},"modified":"2021-02-24T23:10:42","modified_gmt":"2021-02-24T14:10:42","slug":"autoencoder-%eb%aa%a8%eb%8d%b8-%ec%82%ac%ec%9a%a9%ed%95%9c-%ed%83%80%ec%9d%b4%ed%83%80%eb%8b%89-%eb%ac%b8%ec%a0%9c-%ec%a0%91%ea%b7%bc1","status":"publish","type":"post","link":"http:\/\/mukgee.com\/?p=664","title":{"rendered":"AutoEncoder \ubaa8\ub378 \uc0ac\uc6a9\ud55c \ud0c0\uc774\ud0c0\ub2c9 \ubb38\uc81c \uc811\uadfc[1]"},"content":{"rendered":"

\ucf00\uae00\uc5d0\uc11c \uac00\uc7a5 \ucd08\ubcf4\uc801\uc778 \ubb38\uc81c \ud0c0\uc774\ud0c0\ub2c9 \uc0dd\uc874\uc790 \uc608\uce21 \ubb38\uc81c(https:\/\/www.kaggle.com\/c\/titanic\/data<\/a>)\ub97c AutoEncoder\ub85c \uc811\uadfc\ud574\ubcf4\ub824\uace0 \ud569\ub2c8\ub2e4.<\/p>\n

\uc774 \ubb38\uc81c\ub294 \ucf00\uae00\uc5d0\uc11c \uc6cc\ub099 \uc720\uba85\ud55c \ubb38\uc81c\uc774\uae30 \ub54c\ubb38\uc5d0 competition\uc758 \ub9ac\ub354\ubcf4\ub4dc\ub97c \ubcf4\uba74 score\uac00 1\uc785\ub2c8\ub2e4.(\uc815\ud655\ub3c4 100%..)<\/p>\n

AutoEncoder\ub85c \uc811\uadfc\ud558\ub294\uac74 \ub354 \ub192\uc740 score\ub97c \uc704\ud55c \ubaa9\ud45c\uac00 \uc544\ub2cc \ud3c9\uc18c \uad00\uc2ec\uc788\ub358 Task\ub97c AutoEncoder\ub85c \ud574\uacb0\ud560 \uc218 \uc788\uc744\uc9c0\uc5d0 \ub300\ud574 \uc2a4\uc2a4\ub85c \ud655\uc778\ud574\ubcf4\uace0 \uc2f6\uc5b4\uc11c \uc785\ub2c8\ub2e4.<\/p>\n

 <\/p>\n

\uc608\uce21\uc744 \uc911\uc2ec\uc73c\ub85c \uc0dd\uac01\ud558\ub294\uac83\uc774 \uc544\ub2cc AutoEncoder\uc758 \ucc28\uc6d0 \ucd95\uc18c \ub2a5\ub825\uc5d0 \uc911\uc2ec\uc744 \ub450\ub824\uace0 \ud569\ub2c8\ub2e4.<\/p>\n

\uadf8\ub798\uc11c train dataSet\uc5d0 Overfitting \uc5d0 \ub300\ud574\uc11c\ub294 \uc7a0\uae50 \uace0\ubbfc\ud558\uc9c0 \uc54a\uae30\ub85c \ud558\uc8e0.<\/p>\n

\uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\ub97c \uc798 \ucc28\uc6d0 \ucd95\uc18c\ud558\ub294 \ubaa8\ub378\ub9cc \uc6b0\uc120 \ub9cc\ub4e4\uc5b4 \ubd05\uc2dc\ub2e4<\/p>\n

\uadf8\ub9ac\uace0 \ucd95\uc18c\ub41c \ucc28\uc6d0\uc5d0\uc11c knn \uac19\uc740 \uc54c\uace0\ub9ac\uc998\uc73c\ub85c \uc0dd\uc874 \uc5ec\ubd80\ub97c \uc608\uce21\ud574\ubcf4\uae30\ub3c4 \ud558\uace0, Sex\uc640 age \ub4f1\uc73c\ub85c \uc5b4\ub5a4 pclass ticket\uc744 \uad6c\ub9e4\ud558\ub294\uac8c \uc88b\uc744\uc9c0 \ucd94\ucc9c\ud574\ubcf4\ub294 \ubaa8\ub378\ub3c4 \ub9cc\ub4e4\uc5b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n

 <\/p>\n

AutoEncoder\ub85c\ub294 bytenet\uc774\ub77c\ub294 \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud574\ubcf4\ub824\uace0 \ud569\ub2c8\ub2e4.<\/p>\n

 <\/p>\n

\uba3c\uc800 \ucf54\ub7a9\uc5d0\uc11c \ucf00\uae00 \ub370\uc774\ud130\ub97c \ubd88\ub7ec \uc640\uc57c\uaca0\uc8e0<\/p>\n

#\ucf54\ub7a9 \ub77c\uc774\ube0c\ub7ec\ub9ac install\r\n!pip install kaggle\r\nfrom google.colab import files\r\nfiles.upload() #\ucf54\ub7a9\uc758 account\uc5d0\uc11c donwload \ud55c API \ud30c\uc77c update(json\ud30c\uc77c)\r\n\r\n!kaggle competitions download -c titanic #competition\uc758 data \ud0ed\uc5d0\uc11c \ub2e4\uc6b4\ub85c\ub4dc API \uc2e4\ud589\u3141<\/pre>\n
 <\/p>\n
\ud0c0\uc774\ud0c0\ub2c9 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud55c \ub9ce\uc740 notebook\uc744 \ubcf4\uba74 \uc804\ucc98\ub9ac\ub97c \uc704\ud55c \ub9ce\uc740 \uc791\uc5c5\uc774 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ub370\uc774\ud130\uac00 \uc5b4\ub5bb\uac8c \ub9cc\ub4e4\uc5b4\uc84c\ub294\uc9c0 \uc774\ud574\ud558\ub294\uac74 \uadf8\ub9ac\uace0 \uc804\ucc98\ub9ac \uc791\uc5c5\uc740 \ub2f9\uc5f0\ud788 \uc911\uc694\ud558\uc9c0\ub9cc \uc774\ubc88 \ubaa8\ub378\uc740 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\ub97c \uc798 \ucc28\uc6d0 \ucd95\uc18c\ud558\ub294 \ubaa8\ub378\uc744 \ub9cc\ub4dc\ub294\uac8c \ubaa9\ud45c\uc785\ub2c8\ub2e4.<\/p>\n
\uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\ub97c \uadf8\ub0e5 Autoencoder \ubaa8\ub378\uc5d0 \ub123\uc5b4\uc11c 2\ucc28\uc6d0 \uc815\ub3c4\uc758 \ucc28\uc6d0\uc73c\ub85c \ucd95\uc18c\ud569\ub2c8\ub2e4.<\/p>\n
(\ub9ce\uc774 \uc54c\ub824\uc9c4 \ubc29\ubc95 \uc911 name\uc5d0\uc11c title\uc744 \ucd94\ucd9c\ud558\ub294 \uac83\uacfc missing value\ub97c \ucc44\uc6b0\ub294\uac83 \uc815\ub3c4\ub9cc \uc804\ucc98\ub9ac )<\/p>\n
def preprocessing_data(dataset , onehot=False):\r\n  #name\uc5d0\uc11c title \ucd94\ucd9c\r\n  dataset['Title'] = dataset.Name.str.extract(' ([A-Za-z]+)\\.')\r\n\r\n  #missing value \ucc98\ub9ac\r\n  #Index(['PassengerId', 'Survived', 'Pclass', 'Name', 'Sex', 'Age', 'SibSp', 'Parch', 'Ticket', 'Fare', 'Cabin', 'Embarked'] => \uceec\ub7fc\uc740 \ucd1d 12\uac1c\r\n  #print(train.isnull().sum())\r\n  # Age , Cabin, Embarked 3\uac1c \uceec\ub7fc missing\r\n  # Missing value\ub294 \uc784\uc758\uc758 \ub2e4\ub978 \uac12\uc73c\ub85c fill \ucc98\ub9ac\r\n  dataset['Age'].fillna(0 , inplace=True) #Age \ub294 0\uc73c\ub85c\r\n  dataset['Cabin'].fillna('N00' , inplace=True) #Cabin \uc740 N00\r\n  dataset['Embarked'].fillna('N' , inplace=True) #Embarked\ub294 N\uc73c\ub85c\r\n\r\n  #categorical data \ucc98\ub9ac\r\n  # Sex, Ticket, Cabin, Embarked, Title\r\n  cate_col = ['Sex' , 'Ticket' , 'Cabin' , 'Embarked' , 'Title']\r\n  \r\n  for col in cate_col :\r\n    col_label = train[col].astype('category').cat.codes\r\n    if onehot is True:\r\n      num = np.unique(col_label, axis=0).shape[0]\r\n      encoding = np.eye(num)[col_label]\r\n      dataset[col+'_encoding'] = encoding\r\n    else:\r\n      dataset[col+'_encoding'] = col_label\r\n\r\n  return dataset\r\n\r\n\r\ntrain = preprocessing_data(train)\r\ntrain.isnull().sum() #missing value \uc5c6\uc74c\r\ntrain.head()\r\n\r\n#\uc2e4\uc81c \ud559\uc2b5\uc5d0 \uc0ac\uc6a9\ud560 \ub370\uc774\ud130\ub294 sequential \ub370\uc774\ud130\uc640 labeling \ub41c \ub370\uc774\ud130\r\ntrainset = train[['PassengerId' , 'Pclass' , 'Age' , 'SibSp' , 'Parch' , 'Fare' , 'Sex_encoding' , 'Ticket_encoding' , 'Cabin_encoding' , 'Embarked_encoding' , 'Title_encoding']] => 11\uac1c \uceec\ub7fc<\/pre>\n
categorical data\ub97c one-hot encoding\uc744 \uc0dd\uac01\ud574\ubcfc\uc218 \uc788\uc73c\ub098 ticket \ucc98\ub7fc \uc2e4\uc81c \uc758\ubbf8\uac00 \uc788\uc744\uc9c0 \uc54c\uc218 \uc5c6\uc9c0\ub9cc 680\ucc28\uc6d0\uc758 one-hot encoding \uac12\uc744 \uc0dd\uc131\ud574\uc57c\ud558\ub294 \uceec\ub7fc\ub3c4 \uc788\uc5b4\uc11c \uc6b0\uc120\uc740 labeling encoding \ub9cc \uc0ac\uc6a9\ud569\ub2c8\ub2e4.<\/p>\n
(labeling encoding\uc740 pandas\uc5d0\uc11c category type \uc5d0 \ub300\ud574 codes \ub77c\ub294 \uac12\uc73c\ub85c \uc544\uc8fc \uac04\ub2e8\ud558\uac8c \uc0dd\uc131 \uac00\ub2a5\ud569\ub2c8\ub2e4.)<\/p>\n
\ucd5c\uc885 11\uac1c\uc758 \uceec\ub7fc\uc744 \uac00\uc9c0\uace0 \ud559\uc2b5\uc744 \uc2dc\uc791\ud574\ubcfc \uc218 \uc788\uaca0\ub124\uc694.<\/p>\n
 <\/p>\n
\uc6b0\uc120 AutoEncoder \ubaa8\ub378\uc744 \uc0dd\uc131\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc0ac\uc6a9\ud55c \ubaa8\ub378\uc740 Bytenet \uc73c\ub85c Encoder\uc640 Decoder \ubaa8\ub378\uc744 \uc0dd\uc131\ud574\uc11c \uc9c4\ud589\ud569\ub2c8\ub2e4.<\/p>\n
\n
\"\"<\/div>\n<\/div>\n
ByteNet \ubaa8\ub378\uc758 \ucf54\ub4dc\ub294 https:\/\/github.com\/dhpollack\/bytenet.pytorch\/blob\/master\/bytenet\/bytenet_modules.py \uc758 \ucf54\ub4dc\ub97c \uc0ac\uc6a9\ud558\uc600\uc2b5\ub2c8\ub2e4.<\/p>\n
device = torch.device('cuda:0')\r\nencoder = BytenetEncoder(feature_num, max_r, k, num_sets=4).to(device)\r\ndecoder = BytenetDecoder(feature_num , max_r, k, num_sets=4, num_classes=feature_num, use_logsm=False).to(device)\r\nparams = [{\"params\": encoder.parameters()}, {\"params\": decoder.parameters()}]\r\n# criterion = nn.CrossEntropyLoss()\r\n# criterion = torch.nn.Smooth1Loss()\r\ncriterion = torch.nn.L1Loss()\r\noptimizer = torch.optim.Adam(params)\r\n\r\ndataset_sizes = len(train_dataset)<\/pre>\n
Encoder\/Decoder \ubaa8\ub378\uc744 \uc120\uc5b8\ud558\uace0 \ud559\uc2b5 \ud30c\ub77c\ubbf8\ud130\ub97c \ud655\uc778\ud569\ub2c8\ub2e4.<\/p>\n
\uc6d0\ubcf8\uc5d0\uc11c\ub294 ByteNet\uc744 \ubc88\uc5ed\uae30 \ubaa9\uc801\uc73c\ub85c crossEntropyLoss\ub97c \uc0ac\uc6a9\ud558\uc600\uc73c\ub098 Sequential Data\uc5d0\ub294 L1 loss \uac19\uc740 Loss \ub85c reconstruct loss \ub97c \ud655\uc778\ud558\ub294\uac8c \ub9de\uc744\uac83 \uac19\uc544 L1 loss \ub97c \uc0ac\uc6a9\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
(colab\uc5d0\uc11c gpu\ub97c \uc0ac\uc6a9\ud558\uae30 \uc704\ud574\uc11c\ub294 \ub7f0\ud0c0\uc784 \ud658\uacbd\uc744 GPU\ub85c \uc124\uc815\ub9cc \ud558\uba74 \ub429\ub2c8\ub2e4. )<\/p>\n
 <\/p>\n
Deep Learning \ubaa8\ub378\uc744 \ucc98\uc74c \ub9cc\ub4e4\uace0 \uc2e4\ud589 \uc804\uc5d0\ub294 \ud56d\uc0c1 \ucc28\uc6d0\uc744 \ub9de\ucd94\ub294 \uace0\ud1b5\uc2a4\ub7ec\uc6b4 \uc791\uc5c5\uc740 \ud544\uc694\ud55c\uac83 \uac19\uc2b5\ub2c8\ub2e4.<\/p>\n
\ucc28\uc6d0\uc774 \uc548\ub9de\ub294 \ubd80\ubd84\uc744 \uc218\uc815\ud558\uace0 \ub098\uba74 \uac04\ub2e8\ud55c train model \uc744 \ub9cc\ub4e4\uc5b4\uc11c \uc2e4\ud589 \ud558\uba74 \ud559\uc2b5\uc774 \uc9c4\ud589\ub428\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\"\"<\/p>\n
 <\/p>\n
 <\/p>\n
 <\/p>\n
 <\/p>\n
\ucc38\uace0 : https:\/\/medium.com\/hyunjulie\/%EC%BA%90%EA%B8%80%EA%B3%BC-%EA%B5%AC%EA%B8%80-colab-%EC%97%B0%EA%B2%B0%ED%95%B4%EC%A3%BC%EA%B8%B0-6a274f6de81d<\/a><\/p>\n
https:\/\/cyc1am3n.github.io\/2018\/10\/09\/my-first-kaggle-competition_titanic.html<\/a><\/p>\n
https:\/\/towardsdatascience.com\/categorical-encoding-using-label-encoding-and-one-hot-encoder-911ef77fb5bd<\/a><\/p>\n
https:\/\/github.com\/dhpollack\/bytenet.pytorch\/blob\/master\/bytenet\/bytenet_modules.py<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"
\ucf00\uae00\uc5d0\uc11c \uac00\uc7a5 \ucd08\ubcf4\uc801\uc778 \ubb38\uc81c \ud0c0\uc774\ud0c0\ub2c9 \uc0dd\uc874\uc790 \uc608\uce21 \ubb38\uc81c(https:\/\/www.kaggle.com\/c\/titanic\/data)\ub97c AutoEncoder\ub85c \uc811\uadfc\ud574\ubcf4\ub824\uace0 \ud569\ub2c8\ub2e4. \uc774 \ubb38\uc81c\ub294 \ucf00\uae00\uc5d0\uc11c \uc6cc\ub099 \uc720\uba85\ud55c \ubb38\uc81c\uc774\uae30 \ub54c\ubb38\uc5d0 competition\uc758 \ub9ac\ub354\ubcf4\ub4dc\ub97c \ubcf4\uba74 score\uac00 1\uc785\ub2c8\ub2e4.(\uc815\ud655\ub3c4 100%..) AutoEncoder\ub85c…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"spay_email":""},"categories":[3],"tags":[72,73,24],"aioseo_notices":[],"jetpack_featured_media_url":"","_links":{"self":[{"href":"http:\/\/mukgee.com\/index.php?rest_route=\/wp\/v2\/posts\/664"}],"collection":[{"href":"http:\/\/mukgee.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/mukgee.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/mukgee.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/mukgee.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=664"}],"version-history":[{"count":14,"href":"http:\/\/mukgee.com\/index.php?rest_route=\/wp\/v2\/posts\/664\/revisions"}],"predecessor-version":[{"id":680,"href":"http:\/\/mukgee.com\/index.php?rest_route=\/wp\/v2\/posts\/664\/revisions\/680"}],"wp:attachment":[{"href":"http:\/\/mukgee.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=664"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/mukgee.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=664"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/mukgee.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=664"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}