我想知道使用张量流进行回归的好模型(我的火车数据集上有大约200列)
我在网上找到的大多数例子主要用于图像分类,使用卷积网,这对于回归来说是不合适的(我认为)。我找到的极少数回归示例是简单的Xw + b模型,我这样实现:
X = tf.placeholder(tf.float32, [None, train[cols].shape[1]])
Y = tf.placeholder(tf.float32, [None, 1])
W = tf.Variable(tf.zeros([train[cols].shape[1], 1]))
b = tf.Variable(tf.zeros([1]))
preds = tf.matmul(X, W) + b
我尝试了以下代码的几种变体来添加隐藏层,但结果总是最差
X = tf.placeholder(tf.float32, [None, train[cols].shape[1]], name = 'X')
Y = tf.placeholder(tf.float32, [None, 1], name = 'Y')
hidden1 = tf.Variable(tf.random_normal([train[cols].shape[1], 70], stddev=0.01), name = 'hidden1')
b1 = tf.Variable(tf.random_normal([1],stddev=0.01), name ='bias1')
...repeat for other layers....
X = tf.nn.dropout(X, p_keep_input)
h = tf.nn.relu(tf.matmul(X, hidden1) + b1)
h = tf.nn.dropout(h, p_keep_hidden)
h2 = tf.nn.relu(tf.matmul(h, hidden2) +b2)
h2 = tf.nn.dropout(h2, p_keep_hidden)
preds = tf.matmul(h2, w_o)
有什么想法吗? 感谢