| inputlookup  titanic-train.csv

# Ageの欠損値を平均で埋めて、90で割って0-1に収まるようにした
| fit Imputer Age strategy=mean | eval Age = round(Imputed_Age / 90,1)  | fields - Imputed_Age

# Embarked は、欠損値が２つだけだったためDropした。
| search Embarked = * 

# SibSpとParchは数値データだったが、0-1に治めるため文字に変換して、あとで OneHotを行った
| eval SibSp = case(SibSp=0,"S0", SibSp=1,"S1", SibSp=2,"S2", SibSp=3,"S3", SibSp=4,"S4", SibSp=5,"S5",1=1,"S_More")
| eval Parch = case(Parch=0,"P0", Parch=1,"P1", Parch=2,"P2", Parch=3,"P3", Parch=4,"P4", Parch=5,"P5",1=1,"P_More")

# Fareに関しては最大値を計算して、その値でそれぞれ割ることで 0-1の値に収めた
| eventstats max(Fare) as max_Fare
| eval Fare =  round(Fare/max_Fare,1) | eval Fare = "F_" + Fare, Age = "Age_" + Age  | fields - max_Fare

# Name からtitleを抽出し、それを利用した
| rex field=Name "\,\s(?<title>\w+)"  | fields - Name 
| eval title = case(title="Mr","Mr", title="Miss", "Miss", title="Mrs","Mrs",title="Master","Mastar",1=1,"Other")

# Ticketも数字データだけ抽出し、今回はbinを使って分類した
| rex field=Ticket "(?<TicketNo>\d+)$" | fields - Ticket
| bin TicketNo span=100000
| eval TicketNo = case(TicketNo="0-100000","No0", TicketNo="100000-200000", "No100000", TicketNo="200000-300000", "No200000",TicketNo="300000-400000","No300000",1=1,"NoOther")

# CabinはNULL値がほとんどだったが、アルファベット+数値だったため、なんとなく最初のアルファベットが重要だと感じ、アルファベットだけを抽出し利用しました。
| rex field=Cabin "(?<Cabin>\w)" 
| eval Cabin = case(Cabin="T","Cabin_Unknown",Cabin="C","Cabin_C", isnull(Cabin),"Cabin_Unknown",1=1,Cabin)

# OneHotでカテゴリカル変換しました。
| eval {Pclass} = 1, {Sex} = 1, {Embarked} =1, {title}=1, {Cabin} = 1, {TicketNo}=1, {SibSp}=1,{Parch}=1, {Fare} =1, {Age} = 1
| fillnull

# 最後に不要なフィールドは削除
| fields - Pclass, Sex, Embarked, title, count, Cabin, TicketNo, Fare, SibSp, Parch,Age, PassengerId, "Age_0.9"

| fit MLTKContainer mode=stage algo=mytitanic3 Survived from * into app:mytitanic3_model

# mltkc_stage
# this cell is not executed from MLTK and should only be used for staging data into the notebook environment
def stage(name):
    with open("data/"+name+".csv", 'r') as f:
        df = pd.read_csv(f)
    with open("data/"+name+".json", 'r') as f:
        param = json.load(f)
    return df, param

def init(df,param):            
    X = df[param['feature_variables']]
    print("FIT build model with input shape " + str(X.shape))
    input_shape = int(X.shape[1])
    model = keras.Sequential()
    model.add(keras.layers.Dense(input_shape*2, input_dim=input_shape, activation=tf.nn.relu,
                                kernel_initializer='random_uniform'))
    model.add(keras.layers.Dense(32, activation=tf.nn.relu))
    model.add(keras.layers.BatchNormalization())
    model.add(keras.layers.Dropout(0.4))
    model.add(keras.layers.Dense(32, activation=tf.nn.relu))
    model.add(keras.layers.BatchNormalization())
    model.add(keras.layers.Dropout(0.4))
    model.add(keras.layers.Dense(32, activation=tf.nn.relu))
    model.add(keras.layers.BatchNormalization())
    model.add(keras.layers.Dropout(0.4))
    model.add(keras.layers.Dense(32, activation=tf.nn.relu))
    model.add(keras.layers.BatchNormalization())
    model.add(keras.layers.Dropout(0.4))
    model.add(keras.layers.Dense(32, activation=tf.nn.relu))
    model.add(keras.layers.BatchNormalization())
    model.add(keras.layers.Dropout(0.4))
    model.add(keras.layers.Dense(32, activation=tf.nn.relu))
    model.add(keras.layers.Dense(32, activation=tf.nn.relu))
    model.add(keras.layers.Dense(1, activation=tf.nn.sigmoid))
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
    return model

# mltkc_stage_create_model_fit
# returns a fit info json object
def fit(model,df,param):
    returns = {}            
    X = df[param['feature_variables']]
    Y = df[param['target_variables']]
    
    X_train, X_val = np.vsplit(X, [int(X.shape[0] * 0.8)])
    Y_train, Y_val = np.vsplit(Y, [int(Y.shape[0] * 0.8)])
    
    model_epochs = 300
    model_batch_size = 10
    if 'options' in param:
        if 'params' in param['options']:
            if 'epochs' in param['options']['params']:
                model_epochs = int(param['options']['params']['epochs'])
            if 'batch_size' in param['options']['params']:
                model_batch_size = int(param['options']['params']['batch_size'])
    # connect model training to tensorboard
    log_dir="/srv/notebooks/logs/fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
    tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1, embeddings_freq=1)
    
    # Early Stopping
    earlystopping_callback= tf.keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=20, verbose=0, mode='auto')
    
    # run the training
    returns['fit_history'] = model.fit(x=X_train,
                                       y=Y_train, 
                                       verbose=2, 
                                       epochs=model_epochs, 
                                       batch_size=model_batch_size, 
                                       validation_data=(X_val, Y_val),
                                       callbacks=[tensorboard_callback,earlystopping_callback])
    # memorize parameters
    returns['model_epochs'] = model_epochs
    returns['model_batch_size'] = model_batch_size
    returns['model_loss_acc'] = model.evaluate(x = X, y = Y)
    return returns

# mltkc_stage_create_model_apply
def apply(model,df,param):
    X = df[param['feature_variables']]
    y_hat = model.predict(x = X, verbose=0)
    return y_hat

・・・
| fit MLTKContainer algo=mytitanic3 Survived from * into app:mytitanic3_model
| eval predict = if(predicted_Survived > 0.5 , 1,0) 
| score f1_score Survived against predict

・・・
| apply app:mytitanic3_model 
| eval Survived = if(predicted_Survived > 0.5 , 1,0) 
| table PassengerId Survived