python - 用 keras 尝试 Kaggle Titanic .. 得到损失和 valid_loss -0.0000

Question

您好，针对此处发布的问题 (https://www.kaggle.com/c/titanic)，我得到以下代码的奇怪结果 -

from keras.models import Sequential
from keras.layers.core import Dense, Activation, Dropout
from keras.layers.advanced_activations import PReLU, LeakyReLU
from keras.layers.recurrent import SimpleRNN, SimpleDeepRNN
from keras.layers.embeddings import Embedding
from keras.layers.recurrent import LSTM, GRU

import pandas as pd
import numpy as np 
from sklearn import preprocessing

np.random.seed(1919)

### Constants ###
data_folder = "/home/saj1919/Public/Data_Science_Mining_Study/submissions/titanic/data/"
out_folder = "/home/saj1919/Public/Data_Science_Mining_Study/submissions/titanic/output/"
batch_size = 4
nb_epoch = 10

### load train and test ###
train  = pd.read_csv(data_folder+'train.csv', index_col=0)
test  = pd.read_csv(data_folder+'test.csv', index_col=0)
print "Data Read complete"

Y = train.Survived
train.drop('Survived', axis=1, inplace=True)

columns = train.columns
test_ind = test.index

train['Age'] = train['Age'].fillna(train['Age'].mean())
test['Age'] = test['Age'].fillna(test['Age'].mean())
train['Fare'] = train['Fare'].fillna(train['Fare'].mean())
test['Fare'] = test['Fare'].fillna(test['Fare'].mean())

category_index = [0,1,2,4,5,6,8,9]
for i in category_index:
    print str(i)+" : "+columns[i]
    train[columns[i]] = train[columns[i]].fillna('missing')
    test[columns[i]] = test[columns[i]].fillna('missing')

train = np.array(train)
test = np.array(test)

### label encode the categorical variables ###
for i in category_index:
    print str(i)+" : "+str(columns[i])
    lbl = preprocessing.LabelEncoder()
    lbl.fit(list(train[:,i]) + list(test[:,i]))
    train[:,i] = lbl.transform(train[:,i])
    test[:,i] = lbl.transform(test[:,i])

### making data as numpy float ###
train = train.astype(np.float32)
test = test.astype(np.float32)
#Y = np.array(Y).astype(np.int32)

model = Sequential()
model.add(Dense(len(columns), 512))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(512, 1))
model.add(Activation('softmax'))

model.compile(loss='categorical_crossentropy', optimizer="adam")
model.fit(train, Y, nb_epoch=nb_epoch, batch_size=batch_size, validation_split=0.20)
preds = model.predict(test,batch_size=batch_size)

pred_arr = []
for pred in preds:
    pred_arr.append(pred[0])

### Output Results ###
preds = pd.DataFrame({"PassengerId": test_ind, "Survived": pred_arr})
preds = preds.set_index('PassengerId')
preds.to_csv(out_folder+'test.csv')

我得到以下结果:

Train on 712 samples, validate on 179 samples
Epoch 0
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 1
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 2
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 3
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 4
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 5
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 6
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 7
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 8
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000
Epoch 9
712/712 [==============================] - 0s - loss: -0.0000 - val_loss: -0.0000

我正在尝试创建一个简单的 3 层网络。完全基本的代码。在 kaggle 上使用 keras 之前，我已经尝试过这类分类问题。但是这次出现了这个错误。

是否由于数据较少而过度拟合。我缺少什么？有人可以帮忙吗？

Answer 1

旧帖子，但无论如何都要回答，以防其他人尝试使用 Keras 进行泰坦尼克号。

您的网络可能有太多参数和太少正则化(例如丢失)。

在 model.compile 之前调用 model.summary() ，它会显示你的网络有多少参数。在你的两个密集层之间，你应该有 512 X 512 = 262,144 个参数。这对于 762 个示例来说已经很多了。

您可能还想在最后一层使用 sigmoid 激活和 binary_cross 熵损失，因为您只有两个输出类。