About this article

To make FX predictions in deep learning (CNN) From CSV data (date and time, opening price, high price, low price, closing price are described) I would like to generate a large number of chart images.

Deliverables

Create the following chart image from CSV.

About CSV files

The CSV file is an hourly chart of USDJPY. (January 2007-September 2020) If you don't have 1 hour CSV data, please refer to this article . The contents of CSV are as follows. There are more than 80,000 lines.

code

The code is below.

import matplotlib.pylab as plt
import pandas as pd
import numpy as np

def make_sma(parab, arr=None):
    """
Function description: Adds an average moving line to the array received as an argument.
    parab:period
    arr:Array consisting of date and time, open price, high price, low price, close price
    """
    row = arr.shape[1] #Get the number of columns in the array
    arr = np.c_[arr, np.zeros((len(arr),1))] #Add column
    for i in range(parab, len(arr)):
        tmp = arr[i-parab+1:i+1,4].astype(np.float) #Enter the number within the period
        arr[i,row] = np.mean(tmp) #Enter the value of the moving average

    return arr

def dataframe_to_img(chart_range, img_name, df=None):
    """
Function description: Convert DataFrame to image and save.
    chart_range:DataFrame range
    img_name:Image save destination
    df:DataFrame to draw
    """
    df = df[0:chart_range]

    plt.figure()
    df.plot(legend=None) #Delete legend
    plt.axis('off') #Border removal
    plt.tick_params(labelbottom=False,
                    labelleft=False,
                    bottom=False,
                    left=False) #Border removal
    plt.box(False) #Border removal
    plt.savefig(img_name,bbox_inches="tight") #Remove margins and save
    plt.close('all')


#Load csv into array
arr = np.loadtxt(r'CSV file', delimiter=",", skiprows=1, dtype='object') 

#Added technical indicators
arr = make_sma(parab=25, arr=arr)

#Convert to DataFrame
col_name = ['Date',"Open","High","Low","Close","SMA"]
df = pd.DataFrame(arr,columns=col_name)

#Convert to DataFrame and save the image
df = df[df!=0].dropna() #Delete line 0
df = df[['Close','SMA']] #Only the columns to draw on the graph
df = df.astype('float') #Convert to float

chart_range = 360

for i in range(20):
    try:
        img_name = str(i) + '.png' #Image save destination
        dataframe_to_img(chart_range, img_name, df=df[i:chart_range+i])
    except IndexError:
        pass

#DataFrame is also saved because it will be used as a correct label later.
df.to_csv(r'tarintest_labels.csv',encoding='utf_8_sig')

A description of the code. First, load the CSV with numpy.

Then, with ʻarr = make_sma (parab = 25, arr = arr) ` Add a moving average to arr. The period of the moving average is specified by parab. If you want to add long-term lines such as 75 and 200 instead of 25, You can add it by changing the value of parab and adding it.

After adding the technical indicators, With df = pd.DataFrame (arr, columns = col_name) Convert arr to DataFrame.

chart_range = 360 is the range of charts to display. This time it is a 1-hour CSV, so if it is 360, it will be displayed for 360 hours.

Finally, in the for statement, pass the DataFrame to the dataframe_to_img function and save it one by one. dataframe_to_img uses matplotlib. Borders etc. are deleted because they interfere with learning in deep learning.

After outputting the image, save the DataFrame as CSV. This is to use it as a correct label for deep learning.