I tried to predict the up and down of the closing price of Gurunavi's stock price using TensorFlow (progress)
Introduction
-Last time, I'm back again. I'm crazy, but I've remade it quite a bit. ――I have no basic knowledge of deep learning. Please note that I made it only with the desire to move it. ――The development period is about one week. --Python I don't know. This is my first time with TensorFlow. That is the level. --I have no knowledge of stocks. ――The accuracy is not high at all. ――Before the accuracy, I didn't understand how to make the contents, and I remade it over and over again, and once I was exhausted, I only went to study. -GourNavi. ――Since it is in the middle of the process, it may not be fully understood (wrong). ――After you've done it properly, you may organize the pages.
Plagiarism
- Machine Learning with Financial Time Series Data on Google Cloud Platform -[Stock price forecast with TensorFlow] 0 --Try running Google sample code
Description
Predict whether the closing price of Gurunavi's stock price has risen or fallen from the previous day.
Environment
Last time is used.
In addition to this, matplotlib is included to display the graph (it has nothing to do with the actual prediction).
pip install matplotlib
Data collection
The source of the stock price data was obtained from the following. Stock investment memo / stock price database
In addition, we divided the files into 3 files for each brand and combined the files divided for each year. Finally, I adjusted the header column.
It seems that the file cannot be posted, so the following is an example.
Example) Gurunavi.csv
result
Source code
stock_price_prediction.py
#!/usr/local/bin/python
#! -*- coding: utf-8 -*-
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tensorflow as tf
flags = tf.app.flags
FLAGS = flags.FLAGS
#Folder where CSV data is placed
flags.DEFINE_string('csv_dir', '{Directory path}', 'Directory to put the csv data.')
#Folder where learning data is placed
flags.DEFINE_string('train_dir', '{Directory path}', 'Directory to put the training data.')
#Closing column name
flags.DEFINE_string('close_column', 'Close', 'Close column name.')
#AI learning model part(neural network)To create
def inference(num_predictors, num_classes, stock_placeholder):
#Weight with standard deviation 0.Initialize with a normal distribution of 1
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.0001)
return tf.Variable(initial)
#Bias standard deviation 0.Initialize with a normal distribution of 1
def bias_variable(shape):
initial = tf.ones(shape)
return tf.Variable(initial)
with tf.name_scope('fc1') as scope:
weights = weight_variable([num_predictors, num_classes])
biases = bias_variable([num_classes])
#Normalization with softmax function
#Convert the output of the neural network so far to the probability of each label
with tf.name_scope('softmax') as scope:
model = tf.nn.softmax(tf.matmul(stock_placeholder, weights) + biases)
#Probability of each label(Something like?)return it
return model
#Calculate how much "error" there was between the prediction result and the correct answer
def loss(logits, labels):
#Calculation of cross entropy
cross_entropy = -tf.reduce_sum(labels*tf.log(logits))
#Specify to display in TensorBoard
tf.summary.scalar("cross_entropy", cross_entropy)
#Error rate value(cross_entropy)return it
return cross_entropy
#error(loss)Train a learning model designed using error backpropagation based on
def training(labels_placeholder, model):
#Like this function does all that
cost = -tf.reduce_sum(labels_placeholder*tf.log(model))
training_step = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(cost)
return training_step
#Calculate the correct answer rate of the prediction result given by the learning model at inference
def accuracy(model, labels_placeholder):
#Get an array when the predicted (model) and the actual value (actual) match (equal)
#Example result: [1,1,0,1,0]1 is the correct answer
correct_prediction = tf.equal(
tf.argmax(model, 1),
tf.argmax(labels_placeholder, 1)
)
#Result (eg)[1,1,0,1,0]1 is correct) cast to float
#All averages (reduce)_get mean)
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
#Set to display on TensorBoard
tf.summary.scalar("accuracy", accuracy)
return accuracy
if __name__ == '__main__':
stocks = [
'Gurunavi',
'Recruit',
'Kakakucom',
]
closing_data = pd.DataFrame()
for stock in stocks:
#Load CSV into Pandas Dataframe.
data = pd.read_csv(FLAGS.csv_dir + stock + '.csv', index_col='Date').sort_index()
#Normalize the closing price by taking the logarithm of the ratio from one day ago.
#If it is higher than the previous day, it will be positive, and if it is lower, it will be negative.
closing_data[stock] = np.log(data[FLAGS.close_column] / data[FLAGS.close_column].shift())
#Exclude the missing value line.
closing_data = closing_data.dropna()
#Converts a logarithmic return to a flag.
#If positive
closing_data["Gurunavi_Positive"] = 0
closing_data.ix[closing_data["Gurunavi"] >= 0, "Gurunavi_Positive"] = 1
#If negative
closing_data["Gurunavi_Negative"] = 0
closing_data.ix[closing_data["Gurunavi"] < 0, "Gurunavi_Negative"] = 1
training_data = pd.DataFrame(
# column name is "<index>_<day>".
columns= ["Gurunavi_Positive", "Gurunavi_Negative"] + [s + "_1" for s in stocks[1:]]
)
for i in range(7, len(closing_data)):
data = {}
# We will use today's data for positive/negative labels
data["Gurunavi_Positive"] = closing_data["Gurunavi_Positive"].ix[i]
data["Gurunavi_Negative"] = closing_data["Gurunavi_Negative"].ix[i]
# Use yesterday's data for world market data
for col in stocks[1:]:
data[col + "_1"] = closing_data[col].ix[i - 1]
training_data = training_data.append(data, ignore_index=True)
#Data used to make predictions
# Gurunavi_Positive, Gurunavi_Negative
predictors_tf = training_data[training_data.columns[2:]]
#Correct answer data
classes_tf = training_data[training_data.columns[:2]]
#Split training and test data.
training_set_size = int(len(training_data) * 0.8)
test_set_size = len(training_data) - training_set_size
training_predictors_tf = predictors_tf[:training_set_size]
training_classes_tf = classes_tf[:training_set_size]
test_predictors_tf = predictors_tf[training_set_size:]
test_classes_tf = classes_tf[training_set_size:]
#Define variables for the number of predictors and the number of classes to remove the magic number from your code.
num_predictors = len(training_predictors_tf.columns)
num_classes = len(training_classes_tf.columns)
#Specify the scope to be output to the graph of TensorBoard
with tf.Graph().as_default():
#
stock_placeholder = tf.placeholder("float", [None, num_predictors])
#
labels_placeholder = tf.placeholder("float", [None, num_classes])
# inference()To create a model
model = inference(num_predictors, num_classes, stock_placeholder)
# loss()To calculate the loss
loss_value = loss(model, labels_placeholder)
# training()To train and adjust the parameters of the learning model
training_step = training(labels_placeholder, model)
#Accuracy calculation
accuracy = accuracy(model, labels_placeholder)
#Ready to save
saver = tf.train.Saver()
#Creating a Session(TensorFlow calculations must be done in an absolute Session)
sess = tf.Session()
#Variable initialization(Initialize when starting Session)
sess.run(tf.global_variables_initializer())
#TensorBoard display settings(Tensor Board Declarative?)
summary_op = tf.summary.merge_all()
# train_Specify the path to output the TensorBoard log with dir
summary_writer = tf.summary.FileWriter(FLAGS.train_dir, sess.graph)
for step in range(1, 10000):
sess.run(
training_step,
feed_dict={
stock_placeholder: training_predictors_tf.values,
labels_placeholder: training_classes_tf.values.reshape(len(training_classes_tf.values), 2)
}
)
if step % 100 == 0:
train_accuracy = sess.run(
accuracy,
feed_dict={
stock_placeholder: training_predictors_tf.values,
labels_placeholder: training_classes_tf.values.reshape(len(training_classes_tf.values), 2)
}
)
print "step %d, training accuracy %g"%(step, train_accuracy)
#Add a value to be displayed on the TensorBoard after each step
summary_str = sess.run(
summary_op,
feed_dict={
stock_placeholder: training_predictors_tf.values,
labels_placeholder: training_classes_tf.values.reshape(len(training_classes_tf.values), 2)
}
)
summary_writer.add_summary(summary_str, step)
#Display accuracy for test data after training
print "test accuracy %g"%sess.run(
accuracy,
feed_dict={
stock_placeholder: test_predictors_tf.values,
labels_placeholder: test_classes_tf.values.reshape(len(test_classes_tf.values), 2)
}
)
Execution result
python stock_price_prediction.py
2017-06-06 15:26:17.251792: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2017-06-06 15:26:17.251816: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2017-06-06 15:26:17.251824: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2017-06-06 15:26:17.251831: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2017-06-06 15:26:17.251838: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
step 100, training accuracy 0.504931
step 200, training accuracy 0.504931
step 300, training accuracy 0.504931
step 400, training accuracy 0.504931
step 500, training accuracy 0.504931
step 600, training accuracy 0.504931
step 700, training accuracy 0.502959
step 800, training accuracy 0.504931
step 900, training accuracy 0.510848
step 1000, training accuracy 0.510848
step 1100, training accuracy 0.510848
step 1200, training accuracy 0.512821
step 1300, training accuracy 0.510848
step 1400, training accuracy 0.508876
step 1500, training accuracy 0.510848
step 1600, training accuracy 0.510848
step 1700, training accuracy 0.510848
step 1800, training accuracy 0.512821
step 1900, training accuracy 0.510848
step 2000, training accuracy 0.508876
step 2100, training accuracy 0.502959
step 2200, training accuracy 0.499014
step 2300, training accuracy 0.500986
step 2400, training accuracy 0.502959
step 2500, training accuracy 0.504931
step 2600, training accuracy 0.506903
step 2700, training accuracy 0.506903
step 2800, training accuracy 0.514793
step 2900, training accuracy 0.512821
step 3000, training accuracy 0.508876
step 3100, training accuracy 0.504931
step 3200, training accuracy 0.508876
step 3300, training accuracy 0.506903
step 3400, training accuracy 0.510848
step 3500, training accuracy 0.510848
step 3600, training accuracy 0.512821
step 3700, training accuracy 0.512821
step 3800, training accuracy 0.508876
step 3900, training accuracy 0.510848
step 4000, training accuracy 0.512821
step 4100, training accuracy 0.510848
step 4200, training accuracy 0.510848
step 4300, training accuracy 0.512821
step 4400, training accuracy 0.512821
step 4500, training accuracy 0.512821
step 4600, training accuracy 0.512821
step 4700, training accuracy 0.514793
step 4800, training accuracy 0.512821
step 4900, training accuracy 0.512821
step 5000, training accuracy 0.514793
step 5100, training accuracy 0.514793
step 5200, training accuracy 0.514793
step 5300, training accuracy 0.512821
step 5400, training accuracy 0.514793
step 5500, training accuracy 0.514793
step 5600, training accuracy 0.518738
step 5700, training accuracy 0.516765
step 5800, training accuracy 0.518738
step 5900, training accuracy 0.518738
step 6000, training accuracy 0.516765
step 6100, training accuracy 0.514793
step 6200, training accuracy 0.518738
step 6300, training accuracy 0.52071
step 6400, training accuracy 0.518738
step 6500, training accuracy 0.52071
step 6600, training accuracy 0.522682
step 6700, training accuracy 0.522682
step 6800, training accuracy 0.522682
step 6900, training accuracy 0.52071
step 7000, training accuracy 0.52071
step 7100, training accuracy 0.518738
step 7200, training accuracy 0.514793
step 7300, training accuracy 0.516765
step 7400, training accuracy 0.516765
step 7500, training accuracy 0.514793
step 7600, training accuracy 0.512821
step 7700, training accuracy 0.512821
step 7800, training accuracy 0.514793
step 7900, training accuracy 0.514793
step 8000, training accuracy 0.518738
step 8100, training accuracy 0.516765
step 8200, training accuracy 0.516765
step 8300, training accuracy 0.514793
step 8400, training accuracy 0.516765
step 8500, training accuracy 0.518738
step 8600, training accuracy 0.516765
step 8700, training accuracy 0.516765
step 8800, training accuracy 0.516765
step 8900, training accuracy 0.516765
step 9000, training accuracy 0.516765
step 9100, training accuracy 0.516765
step 9200, training accuracy 0.516765
step 9300, training accuracy 0.516765
step 9400, training accuracy 0.516765
step 9500, training accuracy 0.514793
step 9600, training accuracy 0.512821
step 9700, training accuracy 0.512821
step 9800, training accuracy 0.512821
step 9900, training accuracy 0.512821
test accuracy 0.464567
TensorBoard
I wanted to put it out, but I can't put it up because it doesn't seem to be in a normal state.
Supplement
--Accuracy 0.464567 means that it is about 40% that it goes up or down, so it is completely unreliable (?) ――Before you reach this point, learning was a fixed number, but it has finally changed. ――In order to improve accuracy, there is little data, adjustment is necessary, there are few learning elements, or the understanding is wrong.
When you want to put out a graph
# pd.DataFrame()Makes a line graph.
data.plot(figsize = (10, 5), linewidth = 0.5)
#Draw a graph
plt.show()
Recommended Posts