LSTM for time series prediction

Time series prediction Photo by rawpixel.com from Pexels

The idea of using a Neural Network (NN) to predict the stock price movement on the market is as old as Neural nets. Intuitively, it seems difficult to predict the future price movement looking only at its past. There are many tutorials on how to predict the price trend or its power, which simplifies the problem. I’ve decided to try to predict Volume Weighted Average Price with LSTM because it seems challenging and fun.

In this blog post, I am going to train a Long Short Term Memory Neural Network (LSTM) with PyTorch on Bitcoin trading data and use it to predict the price of unseen trading data. I had quite some difficulties with finding intermediate tutorials with a repeatable example of training an LSTM for time series prediction, so I’ve put together a Jupyter notebook to help you to get started.

Loading Necessary Dependencies

Let’s import the libraries that we are going to use for data manipulation, visualization, training the model, etc. We are going to train the LSTM using the PyTorch library.

%matplotlib inline
import glob
import matplotlib
import numpy as np
import pandas as pd
import sklearn
import torch

Loading the Data

We are going to analyze XBTUSD trading data from BitMex. The daily files are…