我是ML初学者。我试图训练一个LSTM模型来预测log2函数，但是结果太不准确了。谁能帮我看看我应该如何改变？非常感谢。
我试图调整超级参数，但仍然没有运气
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
NUM_EPOCH = 1000
HIDDEN_SIZE = 30
NUM_LAYERS = 2
TIMESTEPS = 10
TRAINING_STEPS = 20000
BATCH_SIZE = 20
TRAINING_EXAMPLES = 10000
TESTING_EXAMPLES = 1000
SAMPLE_GAP = 0.01
def generate_data(seq):
X = []
y = []
for i in range(len(seq) - TIMESTEPS):
X.append([seq[i: i + TIMESTEPS]])
y.append([seq[i + TIMESTEPS]])
return np.array(X, dtype=np.float32), np.array(y, dtype=np.float32)
def lstm_model(X, y, is_training):
cell = tf.nn.rnn_cell.MultiRNNCell([tf.nn.rnn_cell.LSTMCell(HIDDEN_SIZE) for _ in range(NUM_LAYERS)])
outputs, _ = tf.nn.dynamic_rnn(cell, X, dtype=tf.float32)
output = outputs[:, -1, :]
predictions = tf.contrib.layers.fully_connected(output, 1, activation_fn=None)
if not is_training:
return predictions, None, None
loss = tf.losses.mean_squared_error(labels=y, predictions=predictions)
train_op = tf.contrib.layers.optimize_loss(
loss, tf.train.get_global_step(), optimizer='Adagrad', learning_rate=0.1)
return predictions, loss, train_op
def train(sess, train_X, train_Y):
ds = tf.data.Dataset.from_tensor_slices((train_X, train_Y))
ds = ds.repeat().shuffle(1000).batch(BATCH_SIZE)
X, y = ds.make_one_shot_iterator().get_next()
losses = np.array([])
with tf.variable_scope('model'):
predictions, loss, train_op = lstm_model(X, y, True)
sess.run(tf.global_variables_initializer())
for i in range(TRAINING_STEPS):
_, l = sess.run([train_op, loss])
losses = np.append(losses, l)
if i % NUM_EPOCH == 0:
print('train step: ' + str(i) + ', loss: ' + str(l))
plt.figure()
plt.plot(losses, label='loss')
plt.legend()
# plt.show()
plt.savefig('./test_outputs/loss.png')
def run_eval(sess, test_X, test_y):
ds = tf.data.Dataset.from_tensor_slices((test_X, test_y))
ds = ds.batch(1)
X, y = ds.make_one_shot_iterator().get_next()
with tf.variable_scope('model', reuse=True):
prediction, _, _ = lstm_model(X, [0, 0], False)
predictions = []
labels = []
for i in range(int(TESTING_EXAMPLES / 2)):
p, l = sess.run([prediction, y])
predictions.append(p)
labels.append(l)
predictions = np.array(predictions).squeeze()
labels = np.array(labels).squeeze()
rmse = np.sqrt(((predictions - labels) ** 2).mean(axis=0))
print('Mean Square Error is: %f' % rmse)
plt.figure()
print(predictions[:15])
print(labels[:15])
plt.plot(predictions, label='predictions')
plt.plot(labels, label='real_val')
plt.legend()
# plt.show()
plt.savefig('./test_outputs/test.png')
test_start = (TRAINING_EXAMPLES + TIMESTEPS) * SAMPLE_GAP + 1
test_end = test_start + (TESTING_EXAMPLES + TIMESTEPS) * SAMPLE_GAP + 1
train_X, train_y = generate_data(np.log2(np.linspace(
SAMPLE_GAP, test_start, TRAINING_EXAMPLES + TIMESTEPS, dtype=np.float32)))
test_X, test_y = generate_data(np.log2(np.linspace(
test_start, test_end, TESTING_EXAMPLES + TIMESTEPS, dtype=np.float32)))
x_val, test_X = np.split(test_X, 2)
y_val, test_y = np.split(test_y, 2)
with tf.Session() as sess:
train(sess, train_X, train_y)
run_eval(sess, test_X, test_y)
培训输出为：
火车步长：0，损失：6.226374 训练步数：1000，损失：0.50559336 训练步数：2000，损失：0.008770825 训练步数：3000，损失：0.006236521 训练步数：4000，损失：0.001877052 训练步数：5000，损失：0.0035174307 训练步数：6000，损失：0.0012169601 训练步数：7000，损失：0.001992577 训练步数：8000，损失：0.00071869115 训练步数：9000，损失：0.0015831447 训练步数：10000，损失：0.0015313074 训练步数：11000，损失：0.0011248669 训练步数：12000，损失：0.0009140003 训练步数：13000，损失：0.0008928403 训练步数：14000，损失：0.00045768628 训练步数：15000，损失：0.000446374 训练步数：16000，损失：0.0010496832 训练步数：17000，损失：0.0008943161 训练步数：18000，损失：0.0010290025 训练步数：19000，损失：0.0005238621 均方误差为：0.052667 [6.6955047 6.6956124 6.69572 6.695827 6.6959357 6.6960425 6.69615 6.6962576 6.6963654 6.696472 6.696579 6.6966867 6.696795 6.696902 6.697009] [6.7375584 6.737707 6.737856 6.7380047 6.738153 6.7383018 6.7384505 6.7385993 6.7387476 6.7388964 6.739045 6.7391934 6.739342 6.7394905 6.7396393] [在此处输入图片描述]
训练输出图像：
