sin波をRNNで推定. r1.0で動作確認済み

参考コードはこちら

コード:nayutaya/tensorflow-rnn-sin/ex1/basic/rnn.py

参考コードがtensorflowのr1.0に対応していなかったので，書き直しました．

r1.01で動作確認しました．

import sys
import yaml # sudo pip3 install pyyaml
import numpy as np
import random

import tensorflow as tf

def make_mini_batch(train_data, size_of_mini_batch, length_of_sequences):
    inputs  = np.empty(0)
    outputs = np.empty(0)
    for _ in range(size_of_mini_batch):
        index   = random.randint(0, len(train_data) - length_of_sequences)
        part    = train_data[index:index + length_of_sequences]
        inputs  = np.append(inputs, part[:, 0])
        outputs = np.append(outputs, part[-1, 1])
    inputs  = inputs.reshape(-1, length_of_sequences, 1)
    outputs = outputs.reshape(-1, 1)
    return (inputs, outputs)

def make_prediction_initial(train_data, index, length_of_sequences):
    return train_data[index:index + length_of_sequences, 0]

train_data_path             = "./train_data/normal.npy"
num_of_input_nodes          = 1
num_of_hidden_nodes         = 2
num_of_output_nodes         = 1
length_of_sequences         = 50
num_of_training_epochs      = 2000
length_of_initial_sequences = 50
num_of_prediction_epochs    = 100
size_of_mini_batch          = 100
learning_rate               = 0.1
forget_bias                 = 1.0
print("train_data_path             = %s" % train_data_path)
print("num_of_input_nodes          = %d" % num_of_input_nodes)
print("num_of_hidden_nodes         = %d" % num_of_hidden_nodes)
print("num_of_output_nodes         = %d" % num_of_output_nodes)
print("length_of_sequences         = %d" % length_of_sequences)
print("num_of_training_epochs      = %d" % num_of_training_epochs)
print("length_of_initial_sequences = %d" % length_of_initial_sequences)
print("num_of_prediction_epochs    = %d" % num_of_prediction_epochs)
print("size_of_mini_batch          = %d" % size_of_mini_batch)
print("learning_rate               = %f" % learning_rate)
print("forget_bias                 = %f" % forget_bias)

train_data = np.load(train_data_path)
print("train_data:", train_data)

train_data_path             = ./train_data/normal.npy
num_of_input_nodes          = 1
num_of_hidden_nodes         = 2
num_of_output_nodes         = 1
length_of_sequences         = 50
num_of_training_epochs      = 2000
length_of_initial_sequences = 50
num_of_prediction_epochs    = 100
size_of_mini_batch          = 100
learning_rate               = 0.100000
forget_bias                 = 1.000000
train_data: [[  0.00000000e+00   1.25333234e-01]
 [  1.25333234e-01   2.48689887e-01]
 [  2.48689887e-01   3.68124553e-01]
 ..., 
 [ -3.68124553e-01  -2.48689887e-01]
 [ -2.48689887e-01  -1.25333234e-01]
 [ -1.25333234e-01   3.92877345e-15]]

# 乱数シードを固定する。
random.seed(0)
np.random.seed(0)
tf.set_random_seed(0)

optimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate)

with tf.Graph().as_default():
    input_ph      = tf.placeholder(tf.float32, [None, length_of_sequences, num_of_input_nodes], name="input")
    supervisor_ph = tf.placeholder(tf.float32, [None, num_of_output_nodes], name="supervisor")
    istate_ph     = tf.placeholder(tf.float32, [None, num_of_hidden_nodes * 2], name="istate") 
    # 1セルあたり2つの値を必要とする。

    with tf.name_scope("inference") as scope:
        weight1_var = tf.Variable(tf.truncated_normal([num_of_input_nodes, num_of_hidden_nodes], stddev=0.1), name="weight1")
        weight2_var = tf.Variable(tf.truncated_normal([num_of_hidden_nodes, num_of_output_nodes], stddev=0.1), name="weight2")
        bias1_var   = tf.Variable(tf.truncated_normal([num_of_hidden_nodes], stddev=0.1), name="bias1")
        bias2_var   = tf.Variable(tf.truncated_normal([num_of_output_nodes], stddev=0.1), name="bias2")

        in1 = tf.transpose(input_ph, [1, 0, 2])         # (batch, sequence, data) -> (sequence, batch, data)
        in2 = tf.reshape(in1, [-1, num_of_input_nodes]) # (sequence, batch, data) -> (sequence * batch, data)
        in3 = tf.matmul(in2, weight1_var) + bias1_var
        
        # r0.1記法
        #in4 = tf.split(0, length_of_sequences, in3)     # sequence * (batch, data)
        # r1.0記法
        in4 = tf.split(in3, length_of_sequences)

        cell = tf.contrib.rnn.BasicLSTMCell(num_of_hidden_nodes, forget_bias=forget_bias, state_is_tuple=False)
        #rnn_output, states_op = rnn.rnn(cell, in4, initial_state=istate_ph) r(0.1)
        rnn_output, states_op = tf.contrib.rnn.static_rnn(cell, in4, initial_state=istate_ph)
        
        output_op
        = tf.matmul(rnn_output[-1], weight2_var) + bias2_var

    with tf.name_scope("loss") as scope:
        square_error = tf.reduce_mean(tf.square(output_op - supervisor_ph))
        loss_op      = square_error
        # tf.scalar_summary("loss", loss_op) # r0.1
        tf.summary.scalar("loss", loss_op) # r.10

    with tf.name_scope("training") as scope:
        training_op = optimizer.minimize(loss_op)

    # summary_op = tf.merge_all_summaries() #0.1
    summary_op = tf.summary.merge_all() # 1.0
    
    init = tf.initialize_all_variables()

    with tf.Session() as sess:
        saver = tf.train.Saver()
        # summary_writer = tf.train.SummaryWriter("data", graph=sess.graph) # r0.1
        summary_writer = tf.summary.FileWriter("data", graph=sess.graph) # r1.0

        sess.run(init)

        for epoch in range(num_of_training_epochs):
            inputs, supervisors = make_mini_batch(train_data, size_of_mini_batch, length_of_sequences)

            train_dict = {
                input_ph:      inputs,
                supervisor_ph: supervisors,
                istate_ph:     np.zeros((size_of_mini_batch, num_of_hidden_nodes * 2)),
            }
            sess.run(training_op, feed_dict=train_dict)

            if (epoch + 1) % 10 == 0:
                summary_str, train_loss = sess.run([summary_op, loss_op], feed_dict=train_dict)
                summary_writer.add_summary(summary_str, epoch)
                print("train#%d, train loss: %e" % (epoch + 1, train_loss))

        inputs  = make_prediction_initial(train_data, 0, length_of_initial_sequences)
        outputs = np.empty(0)
        states  = np.zeros((num_of_hidden_nodes * 2)),

        print("initial:", inputs)
        np.save("initial.npy", inputs)

        for epoch in range(num_of_prediction_epochs):
            pred_dict = {
                input_ph:  inputs.reshape((1, length_of_sequences, 1)),
                istate_ph: states,
            }
            output, states = sess.run([output_op, states_op], feed_dict=pred_dict)
            print("prediction#%d, output: %f" % (epoch + 1, output))

            inputs  = np.delete(inputs, 0)
            inputs  = np.append(inputs, output)
            outputs = np.append(outputs, output)

        print("outputs:", outputs)
        np.save("output.npy", outputs)

        saver.save(sess, "data/model")

WARNING:tensorflow:<tensorflow.contrib.rnn.python.ops.core_rnn_cell_impl.BasicLSTMCell object at 0x00000000118F9160>: Using a concatenated state is slower and will soon be deprecated.  Use state_is_tuple=True.
WARNING:tensorflow:From <ipython-input-15-d5dd1fd24074>:42: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.
Instructions for updating:
Use `tf.global_variables_initializer` instead.
train#10, train loss: 5.229673e-01
train#20, train loss: 5.009690e-01
train#30, train loss: 5.198909e-01
train#40, train loss: 5.346115e-01
train#50, train loss: 4.447417e-01
train#60, train loss: 4.501579e-01
train#70, train loss: 4.537308e-01
train#80, train loss: 3.279701e-01
prediction#87, output: -0.983231
prediction#88, output: -0.955930
prediction#89, output: -0.914552
prediction#90, output: -0.859774
prediction#91, output: -0.791508
prediction#92, output: -0.709254
prediction#93, output: -0.612653
prediction#94, output: -0.502259
prediction#95, output: -0.380304
prediction#96, output: -0.250905
prediction#97, output: -0.119140
prediction#98, output: 0.010653
prediction#99, output: 0.136095
prediction#100, output: 0.256669
outputs: [ 0.00926255  0.14518531  0.26629266  0.3826645   0.4946214   0.60127538
  0.70039088  0.78870726  0.86278987  0.92006391  0.95948833  0.98155129
  0.98773634  0.97983921  0.95942783  0.92752028  0.88444293  0.82979822
  0.76252556  0.68110597  0.58405989  0.47093415  0.34376413  0.20819654
  0.07255314 -0.05579655 -0.17396016 -0.28367797 -0.38881654 -0.49280256
 -0.59695822 -0.69951463 -0.79543906 -0.87775397 -0.94028378 -0.98007071
 -0.99764407 -0.99550325 -0.97644424 -0.94260907 -0.89518976 -0.83448744
 -0.76016831 -0.67169166 -0.56896263 -0.45314986 -0.32730505 -0.19613414
 -0.06463058  0.06360264  0.18705504  0.30573574  0.42002755  0.52970022
  0.63335198  0.72833896  0.81128317  0.87901324  0.92949694  0.96226823
  0.97820294  0.97890699  0.96607846  0.94105148  0.90454733  0.85657418
  0.79643178  0.72282809  0.63420713  0.52946943  0.40922943  0.27726871
  0.14084016  0.00827865 -0.11505078 -0.22876213 -0.33597133 -0.44053042
 -0.54490018 -0.64883077 -0.74872804 -0.83822566 -0.91042352 -0.9607482
 -0.98827887 -0.99484038 -0.98323065 -0.95593017 -0.91455233 -0.85977364
 -0.7915076  -0.70925373 -0.6126532  -0.50225872 -0.38030428 -0.25090536
 -0.11914031  0.01065348  0.13609533  0.25666925]

付録

記法の修正

・Cell関連

# r0.1
cell = tf.nn.rnn.LSTMCell()
# r1.0
cell = tf.contrib.rnn.BasicRNNCell()

・rnn.rnn() => tf.contrib.rnn.static_rnn()

# r0.1
from tensorflow.models.rnn import rnn, rnn_cell
rnn_output, states_op = rnn.rnn(cell, in4, initial_state=istate_ph) r(0.1)
# r1.0
rnn_output, states_op = tf.contrib.rnn.static_rnn(cell, in4, initial_state=istate_ph)

・split

# r0.1記法
in4 = tf.split(0, length_of_sequences, in3)     # sequence * (batch, data)
# r1.0記法
in4 = tf.split(in3, length_of_sequences)

・summary

# tf.scalar_summary("loss", loss_op) # r0.1
tf.summary.scalar("loss", loss_op) # r.10

# summary_op = tf.merge_all_summaries() #0.1
summary_op = tf.summary.merge_all() # 1.0

# summary_writer = tf.train.SummaryWriter("data", graph=sess.graph) # r0.1
summary_writer = tf.summary.FileWriter("data", graph=sess.graph) # r1.0