Titanic - Kaggle competition

First try - Score：0.74162 / Rate: 14044

use Sex,Age,family=SibSp+Parch+1,SocialClass=Fare/Pclass,Cabin as feature

Analyse: high variance and overfitting !

Second try - Score：0.76794 / Rate: 10523

Based on first-try's Analyse, changes are below:

1. use less features
2. use feature-regularizer

use Sex,Age,family=SibSp+Parch+1,SocialClass=Fare/Pclass as feature

use

model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(64, activation='relu', kernel_regularizer=L2(0.001)),
    tf.keras.layers.Dense(16, activation='relu', kernel_regularizer=L2(0.001)),
    tf.keras.layers.Dense(1, activation='sigmoid', kernel_regularizer=L2(0.001)),
])

instead of

model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(64, activation='relu'),
    tf.keras.layers.Dense(16, activation='relu'),
    tf.keras.layers.Dense(1, activation='sigmoid'),
])