titanic_keras.py

from keras.optimizers import SGD, Adam
from keras.layers import Dense, Activation, Dropout
from keras.models import Sequential
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import LabelEncoder
import numpy as np
import pandas as pd


def normalize_data(data):
    not_concerned_columns = ["PassengerId", "Name",
                             "Ticket", "Fare", "Cabin", "Embarked"]
    data = data.drop(not_concerned_columns, axis=1)
    data = data.dropna()

    # normalize
    dummy_columns = ["Pclass"]
    for column in dummy_columns:
        data = pd.concat([data, pd.get_dummies(
            data[column], prefix=column)], axis=1)
        data = data.drop(column, axis=1)

    # normalize Label:Sex to int
    le = LabelEncoder()
    le.fit(["male", "female"])

    data["Sex"] = le.transform(data["Sex"])

    # normalize Age
    ss = StandardScaler()
    data["Age"] = ss.fit_transform(data["Age"].values.reshape(-1, 1))
    return data


def split_train_and_test(data, rate=0.8):
    data_y = data["Survived"]
    data_x = data.drop(["Survived"], axis=1)

    train_valid_split_idx = int(len(data_x) * rate)
    train_x = data_x[:train_valid_split_idx]
    train_y = data_y[:train_valid_split_idx]

    valid_test_split_idx = (len(data_x) - train_valid_split_idx) // 2
    test_x = data_x[train_valid_split_idx + valid_test_split_idx:]
    test_y = data_y[train_valid_split_idx + valid_test_split_idx:]

    return train_x.values, train_y.values.reshape(-1, 1), test_x.values, test_y.values.reshape(-1, 1)


def build_model(input_dim):
    model = Sequential()
    model.add(Dense(20, input_dim=input_dim))
    model.add(Activation('relu'))
    model.add(Dense(1, input_dim=20))
    model.add(Activation('sigmoid'))
    model.compile(optimizer=SGD(lr=0.01),
                  loss='binary_crossentropy',
                  metrics=['accuracy'])
    return model


# load data
train_data = pd.read_csv("data/train.csv")
normalized_data = normalize_data(train_data)
train_x, train_y, test_x, test_y = split_train_and_test(normalized_data, 0.8)

model = build_model(train_x.shape[1])

# train
model.fit(train_x, train_y, nb_epoch=120, batch_size=16)

# test
[loss, accuracy] = model.evaluate(test_x, test_y)
print("loss:{0} -- accuracy:{1}".format(loss, accuracy))