用TensorFlow完成Kaggle Titanic竞赛

Kaggle Titanic竞赛是根据信息，预测是否能生还，是一个典型的二分类问题。

第一步，查看数据，根据直觉来洞察数据

import pandas as pd 
dataset = pd.read_csv("train.csv")
print(dataset.head())
print(dataset.isna().sum())

PassengerId 0
Survived 0
Pclass 0
Name 0
Sex 0
Age 177
SibSp 0
Parch 0
Ticket 0
Fare 0
Cabin 687
Embarked 2

Cabin缺失的数据很多，Cabin number(客舱号)跟是否能生还，没有太大联系，所以该列数据不作为特征。Embarked，作为登船的港口，也没太大联系，不作为特征。Ticket(票号)与Fare(票价)重复，Fare(票价)作为特征更好。
姓名也没有什么直观联系，不作为特征；PassengerID，也没有什么直观联系。

年龄有177个缺失，后面做两种方案，一种带年龄；一种不带年龄。
最终可得特征向量:
第一种： ['pclass','sex','sibsp','parch','fare']
第二种： ['pclass','sex','age','sibsp','parch','fare']

第二步，数据清洗：The 4 C's of Data Cleaning: Correcting, Completing, Creating, and Converting

• Correcting，把不合理的数据更正
• Completing, 把缺失的数据补充完整。定量数据的基本方法是使用平均值、中位数或平均值 + 随机标准差进行插补，例如，年龄将使用中位数进行估算。
• Creating，特征工程是指我们使用现有特征来创建新特征
• Converting，消除单位的影响，归一化等

第三步，这是一个二分类问题，手动提取了特征向量后，就可以用keras建模并训练了，Python代码如下：

import pandas as pd 
import numpy as np 
import tensorflow as tf
from tensorflow.python.keras import activations 
import matplotlib.pylab as plt 
from tensorflow.keras import regularizers

raw_train_dataset = pd.read_csv("train.csv")
raw_test_dataset = pd.read_csv("test.csv")

features_test = ['PassengerId','Pclass','Sex','Fare','Age']
features_train = features_test + ['Survived']

def preprocess(raw_dataset, features, train=True):
    if 'Age' in features:
        raw_dataset['Age'].fillna(raw_dataset['Age'].median(), inplace=True)
    raw_dataset['Fare'].fillna(raw_dataset['Fare'].median(),inplace=True)

    dataset = raw_dataset[features]
    dataset.replace('male',1,inplace=True)
    dataset.replace('female',0,inplace=True)
    dataset_withoutna = dataset.dropna()

    if train:
        labels = dataset_withoutna['Survived']
        dataset_withoutna.pop('PassengerId')
        dataset_withoutna.pop('Survived')
        #print(dataset_withoutna.head())
        train_stats = dataset_withoutna.describe()
        train_stats = train_stats.transpose()
        normed_train_data = (dataset_withoutna - train_stats['mean']) / train_stats['std']
        #print(normed_train_data.head())
        return np.array(normed_train_data), np.array(labels)
    else:
        labels = dataset.pop('PassengerId')
        dataset.fillna(0, inplace=True)
        train_stats = dataset.describe()
        train_stats = train_stats.transpose()
        normed_test_data = (dataset - train_stats['mean']) / train_stats['std']
        return np.array(normed_test_data), np.array(labels)

train_dataset, labels = preprocess(raw_train_dataset, features_train)
print(train_dataset.shape, labels.shape)

test_dataset, passenger_id = preprocess(raw_test_dataset, features_test, train=False)
print(test_dataset.shape, passenger_id.shape)


model = tf.keras.Sequential(
    [
        tf.keras.layers.Dense(64, activation='relu', input_shape=(train_dataset.shape[1],), kernel_regularizer=regularizers.l2(0.001)),
        tf.keras.layers.Dense(32, activation='relu',kernel_regularizer=regularizers.l2(0.001)),
        tf.keras.layers.Dense(16, activation='relu'),
        tf.keras.layers.Dense(1, name='prediction')
    ]
)

base_lr = 0.001
model.compile(
    optimizer=tf.keras.optimizers.Adam(learning_rate=base_lr), 
    loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
    metrics=['accuracy'])

#early_stop = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=10)

history = model.fit(
    train_dataset,labels,
    epochs=400,
    validation_split=0.2,
    batch_size=32,
    verbose=0,
    #callbacks=[early_stop]
)

def plot_history(history):
    hist = pd.DataFrame(history.history)
    hist['epoch'] = history.epoch

    plt.figure()
    plt.xlabel('Num of Epochs')
    plt.ylabel('value')
    plt.plot(hist['epoch'], hist['loss'],
           label='Loss')
    plt.plot(hist['epoch'], hist['val_loss'],
           label = 'val_loss')
    plt.ylim([0,5])
    plt.legend()
    plt.show()

loss, accuracy = model.evaluate(train_dataset, labels, verbose=2)
print("Accuracy:", accuracy)
plot_history(history)

predictions = model.predict(test_dataset)
predictions = (tf.sigmoid(predictions).numpy().flatten() > 0.5).astype(int)
print(predictions.shape, predictions)

output = pd.DataFrame({'PassengerId':passenger_id, 'Survived':predictions})
output.to_csv("submission.csv", index=False)
print("Your submisstion was successfully saved!")