Question

我是机器学习和尝试TensorFlow的新手。为了我的学习，我试图构建一个二进制分类器以在经典的《泰坦尼克号》生存数据上运行。我有几个问题需要解决，我们将不胜感激。

我尝试了两种方法来计算准确度，并且彼此不同意。请参阅tf.name_scope（“ eval”）部分。
如何最好使用内置的tf.metrics.auc（）函数计算准确度的AUC？
我能够保存模型，但是如何加载并进行预测？

代码如下，我省略了数据格式化功能。

# Split the data into X and y.
# Read data
df_train = pd.read_csv("train.csv", index_col=0)
df = format_data(df_train)
df_X = df.drop(labels="Survived", axis=1)
df_X = df_X.astype('float32') # Float32 data type in all the columns.
df_y = pd.DataFrame(data=df.Survived, columns=["Survived"]) # Float32 datatype in all the columns.
df_y = df_y.astype('float32')

X_train, X_test, y_train, y_test = train_test_split(df_X, df_y, test_size=0.3, random_state=100)

# Tensorflow graph construction phase.
tf.reset_default_graph() # Reset default graph if you want to execute this cell multiple times. 

n_inputs = df_X.shape[1]
n_hidden1 = 10
n_hidden2 = 5
n_outputs = df_y.shape[1]
X = tf.placeholder(tf.float32, shape=(None,n_inputs), name="X")
y = tf.placeholder(tf.float32, shape=(None), name="y")

# Define network architecture
hidden1 = tf.layers.dense(X, n_hidden1, activation=tf.nn.relu,
                          kernel_initializer = tf.contrib.layers.xavier_initializer(),
                          kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=1.0), 
                          name="hidden1")

hidden2 = tf.layers.dense(hidden1, n_hidden2, activation=tf.nn.relu,
                          kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=1.0), 
                          name="hidden2")

logits = tf.layers.dense(hidden1, n_outputs, name="outputs")

with tf.name_scope("loss"):
    loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=logits, labels=y)

with tf.name_scope("train"):
    learning_rate = 0.01
    trainer = tf.train.AdamOptimizer(learning_rate).minimize(loss)

with tf.name_scope("eval"):
    # Define the accuracy
    #The default threshold is 0.5, rounded off directly
    prediction = tf.round(tf.sigmoid(logits))
    # Bool into float32 type
    correct = tf.cast(tf.equal(prediction, y), dtype=tf.float32)
    # Average
    accuracy = tf.reduce_mean(correct)

    # Using tf.metrics.accuracy method to calculate accuracy
    acc, acc_update = tf.metrics.accuracy(labels=y, predictions=prediction)

# Assign variable initializer
init = tf.global_variables_initializer()
init_local = tf.local_variables_initializer()

# Assign model saver
saver = tf.train.Saver()

# TensorFlow execution phase.
n_epochs = 1000

with tf.Session() as sess:
    init.run()
    init_local.run()
    for epoch in range(n_epochs):
        sess.run(trainer, feed_dict={X:X_train, y:y_train})
        acc_train = accuracy.eval(feed_dict={X:X_train, y:y_train})
        acc_test = accuracy.eval(feed_dict={X:X_test, y:y_test})
        acc1_train = acc_update.eval(feed_dict={X:X_train, y:y_train})
        acc1_test = acc_update.eval(feed_dict={X:X_test, y:y_test})
    print("Training and test accuracy at", epoch, "epoch:", acc_train, acc_test)
    print("Training and test accuracy at", epoch, "epoch:", acc1_train, acc1_test)
    save_path = saver.save(sess, "./titanic.ckpt")

这是我使用的两种精度方法的输出。

Training and test accuracy at 999 epoch: 0.9069021 0.8507463
Training and test accuracy at 999 epoch: 0.8738184 0.8738114

Tensorflow：二进制分类器的测量精度和AUC

0 个答案: