我正在尝试使用1D-CNN函数模型进行库存移动预测，如何减少验证时的过拟合？
使用股票市场价格的开市价，收市价，最高价，最低价和交易量的历史记录输入模型。因此，输入将为矩阵 WINDOW_SIZE X 5 。输出将检查最后一天的下一个低价，如果上涨，它将下跌，分别标记为“上涨”和“下跌”。我使用此paper的引用创建了模型。
在纸张状态下，他正在使用6层转化，而最后3层的每层过滤大小分别为32、64、128和256。 我已经尝试过了：
这是我一直在使用的数据集csv file
这是指向full code
的链接这里是使用tensorflow的模型：
WINDOW = 30
n_channels = 5
n_classes = 2
batch_size = 100 # Batch size
learning_rate = 0.0001
epochs = 10000
graph = tf.Graph()
# Construct placeholders
with graph.as_default():
inputs_ = tf.placeholder(tf.float32, [None, WINDOW, n_channels], name = 'inputs')
labels_ = tf.placeholder(tf.float32, [None, n_classes], name = 'labels')
keep_prob_ = tf.placeholder(tf.float32, name = 'keep')
learning_rate_ = tf.placeholder(tf.float32, name = 'learning_rate')
with graph.as_default():
# Layer-1
conv = tf.layers.conv1d(inputs=inputs_, filters=32, kernel_size=3, strides=1, padding='valid', activation=tf.nn.relu)
conv = tf.layers.batch_normalization(conv)
# Layer-2
conv = tf.layers.conv1d(inputs=conv, filters=64, kernel_size=3, strides=1, padding='valid')
conv = tf.nn.leaky_relu(conv, alpha=0.01)
conv = tf.layers.batch_normalization(conv)
# Layer-3
conv = tf.layers.conv1d(inputs=conv, filters=128, kernel_size=3, strides=1, padding='valid')
conv = tf.nn.leaky_relu(conv, alpha=0.01)
conv = tf.layers.batch_normalization(conv)
# Layer-4
conv = tf.layers.conv1d(inputs=conv, filters=256, kernel_size=3, strides=1, padding='valid')
conv = tf.nn.leaky_relu(conv, alpha=0.01)
conv = tf.layers.batch_normalization(conv)
# Layer-5
conv = tf.layers.conv1d(inputs=conv, filters=256, kernel_size=3, strides=1, padding='valid')
conv = tf.nn.leaky_relu(conv, alpha=0.01)
conv = tf.layers.batch_normalization(conv)
# Layer-6
conv = tf.layers.conv1d(inputs=conv, filters=256, kernel_size=3, strides=1, padding='valid')
conv = tf.nn.leaky_relu(conv, alpha=0.01)
conv = tf.layers.batch_normalization(conv)
with graph.as_default():
# Flatten and add dropout
flat = tf.layers.flatten(conv)
flat = tf.nn.dropout(flat, keep_prob=keep_prob_)
dense = tf.layers.dense(inputs=flat, units=256)
dense = tf.nn.dropout(dense, rate=1-keep_prob_)
dense = tf.layers.dense(inputs=dense, units=256)
dense = tf.nn.dropout(dense, rate=1-keep_prob_)
# Predictions
logits = tf.layers.dense(flat, n_classes)
# Cost function and optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=labels_))
optimizer = tf.train.AdamOptimizer(learning_rate_).minimize(cost)
# Accuracy
correct_pred = tf.equal(tf.argmax(logits, 1), tf.argmax(labels_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32), name='accuracy')
validation_acc = []
validation_loss = []
train_acc = []
train_loss = []
with graph.as_default():
saver = tf.train.Saver()
with tf.Session(graph=graph) as sess:
sess.run(tf.global_variables_initializer())
iteration = 1
for e in range(epochs):
# Loop over batches
for x,y in get_batches(X_tr, y_tr, batch_size):
# Feed dictionary
feed = {inputs_ : x, labels_ : y, keep_prob_ : 0.2, learning_rate_ : learning_rate}
# Loss
loss, _ , acc = sess.run([cost, optimizer, accuracy], feed_dict = feed)
train_acc.append(acc)
train_loss.append(loss)
# Print at each 5 iters
if (iteration % 50 == 0):
print("Epoch: {}/{}".format(e, epochs),
"Iteration: {:d}".format(iteration),
"Train loss: {:6f}".format(loss),
"Train acc: {:.6f}".format(acc))
# Compute validation loss at every 10 iterations
if (iteration%100 == 0):
val_acc_ = []
val_loss_ = []
for x_v, y_v in get_batches(X_vld, y_vld, batch_size):
# Feed
feed = {inputs_ : x_v, labels_ : y_v, keep_prob_ : 1.0}
# Loss
loss_v, acc_v = sess.run([cost, accuracy], feed_dict = feed)
val_acc_.append(acc_v)
val_loss_.append(loss_v)
# Print info
print("Epoch: {}/{}".format(e, epochs),
"Iteration: {:d}".format(iteration),
"Validation loss: {:6f}".format(np.mean(val_loss_)),
"Validation acc: {:.6f}".format(np.mean(val_acc_)))
# Store
validation_acc.append(np.mean(val_acc_))
validation_loss.append(np.mean(val_loss_))
# Iterate
iteration += 1
saver.save(sess,"/checkpoints-cnn/har.ckpt")
这是我的结果
