search

dolremi / Titanic-kaggle

Titanic project for Kaggle
0 stars 1 forks source link

Enh: try with NN models #1

Open jmzhao opened 8 years ago

jmzhao commented 8 years ago

I am going to apply some multi-layer neural networks on some most significant features to see how it works out, with no or minimal feature engineering.

jmzhao commented 8 years ago

Start with features ['Sex', 'Age', 'Age_invalid', 'Pclass', 'Fare_invalid', 'LogFare']. Got training accuracy 0.81 after 50 eppch. Model is MLP with 3 hidden layers: Input(6), Dense(16, tanh), Dense(32, tanh), Dropout(0.25), Dense(32, tanh), Dropout(0.5), Dense(1, sigmod)