Professional Documents
Culture Documents
io
Learning
If you were to pick the three most ridiculous fads of 2017, they would
definitely be fidget spinners (are they still cool? Do kids still use that word
“cool”?), artificial intelligence and, yes, cryptocurrencies. Joking aside, I’m
actually impressed by the underlying concept and I’m quite bullish on the
long term prospects of this disruptive technology. But enough about fidget
spinners!!! I’m actually not a hodler of any cryptos. So, while I may not
have a ticket to the moon, I can at least get on board the hype train by
successfully predicting the price of cryptos by harnessing deep learning,
machine learning and artificial intelligence (yes, all of them!).
We’re going to employ a Long Short Term Memory (LSTM) model; it’s a
particular type of deep learning model that is well suited to time series data
(or any data with temporal/spatial/structural order e.g. movies, sentences,
etc.). If you wish to truly understand the underlying theory (what kind of
crypto enthusiast are you?), then I’d recommend this blog or this blog or
the original (white)paper. As I’m shamelessly trying to appeal to a wider
non-machine learning audience, I’ll keep the code to a minimum. There’s a
Jupyter (Python) notebook available here, if you want to play around with
the data or build your own models. Let’s get started!
Data
Before we build the model, we need to obtain some data for it. There’s a
dataset on Kaggle that details minute by minute Bitcoin prices (plus some
other factors) for the last few years (featured on that other blog post). Over
this timescale, noise could overwhelm the signal, so we’ll opt for daily
prices. The issue here is that we may have not su cient data (we’ll have
hundreds of rows rather than thousands or millions). In deep learning, no
model can overcome a severe lack of data. I also don’t want to rely on static
files, as that’ll complicate the process of updating the model in the future
with new data. Instead, we’ll aim to pull data from websites and APIs.
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 1/11
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
import pandas as pd
import time
import seaborn as sns
import matplotlib.pyplot as plt
import datetime
import numpy as np
# get market info for bitcoin from the start of 2016 to the cu
bitcoin_market_info = pd.read_html("https://coinmarketcap.com/
# convert the date string to the correct date format
bitcoin_market_info = bitcoin_market_info.assign(Date=pd.to_da
# when Volume is equal to '-' convert it to 0
bitcoin_market_info.loc[bitcoin_market_info['Volume']=="-",'Vo
# convert to int
bitcoin_market_info['Volume'] = bitcoin_market_info['Volume']
# look at the first few rows
bitcoin_market_info.head()
To explain what’s just happened, we’ve loaded some python packages and
then imported the table that you see on this site. With a little bit of data
cleaning, we arrive at the above table. We also do the same thing for ether
by simply replacing ‘bitcoin’ with ‘ethereum’ in the url (code omitted).
To prove that the data is accurate, we can plot the price and volume of both
cryptos over time.
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 2/11
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
We have some data, so now we need to build a model. In deep learning, the
data is typically split into training and test sets. The model is built on the
training set and subsequently evaluated on the unseen test set. In time
series models, we generally train on one period of time and then test on
another separate period. Rather arbitrarily, I’ll set the cut-o date to June
1st 2017 (i.e. model will be trained on data before that date and assessed on
data after it).
You can see that the training period mostly consists of periods when
cryptos were relatively cheaper. As such, the training data may not be
representative of the test data, undermining the model’s ability to
generalise to unseen data (you could try to make your data stationary-
discussed here). But why let negative realities get in the way of baseless
optimism? Before we take our deep artificially intelligent machine learning
model to the moon, it’s worth discussing a simpler model. The most basic
model is to set tomorrow’s price equal to today’s price (which we’ll crudely
call a lag model). This is how we’d define such a model in mathematical
terms:
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 3/11
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
Extending this trivial lag model, stock prices are commonly treated as
random walks, which can be defined in these mathematical terms:
We’ll determine μ and σ from the training sets and apply the random walk
model to the Bitcoin and Ethereum test sets.
Wow! Look at those prediction lines. Apart from a few kinks, it broadly
tracks the actual closing price for each coin. It even captures the eth rises
(and subsequent falls) in mid-June and late August. At this stage, if I was to
announce the launch of sheehanCoin, I’m sure that ICO would stupidly
over-subscribed. As pointed out on that other blog, models that only make
predictions one point into the future are often misleadingly accurate, as
errors aren’t carried over to subsequent predictions. No matter how large
the error, it’s essentially reset at each time point, as the model is fed the
true price. The Bitcoin random walk is particularly deceptive, as the scale of
the y-axis is quite wide, making the prediction line appear quite smooth.
P redP rice t = P redP rice t−1 ∗ ϵ, ϵ ∼ N (μ, σ) & P redP rice 0 = P rice 0
Let’s get our random walk model to predict the closing prices over the total
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 4/11
test set.
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
The model predictions are extremely sensitive to the random seed. I’ve
selected one where the full interval random walk looks almost decent for
Ethereum. In the accompanying Jupyter notebook, you can interactively
play around with the seed value below to see how badly it can perform.
Notice how the single point random walk always looks quite accurate, even
though there’s no real substance behind it. Hopefully, you’ll be more
suspicious of any blog that claims to accurately predict prices. I probably
shouldn’t worry; it’s not like crypto fans to be seduced by slick marketing
claims.
Like I said, if you’re interested in the theory behind LSTMs, then I’ll refer
you to this, this and this. Luckily, we don’t need to build the network from
scratch (or even understand it), there exists packages that include standard
implementations of various deep learning algorithms (e.g. TensorFlow,
Keras, PyTorch, etc.). I’ll opt for Keras, as I find it the most intuitive for
non-experts. If you’re not familiar with Keras, then check out my previous
tutorial.
model_data.head()
I’ve created a new data frame called model_data. I’ve removed some of the
previous columns (open price, daily highs and lows) and reformulated some
new ones. close_off_high represents the gap between the closing price
and price high for that day, where values of -1 and 1 mean the closing price
was equal to the daily low or daily high, respectively. The volatility
columns are simply the di erence between high and low price divided by
the opening price. You may also notice that model_data is arranged in
order of earliest to latest. We don’t actually need the date column anymore,
as that information won’t be fed into the model.
Our LSTM model will use previous data (both bitcoin and eth) to predict the
next day’s closing price of a specific coin. We must decide how many
previous days it will have access to. Again, it’s rather arbitrary, but I’ll opt
for 10 days, as it’s a nice round number. We build little data frames
consisting of 10 consecutive days of data (called windows), so the first
window will consist of the 0-9th rows of the training set (Python is zero-
indexed), the second will be the rows 1-10, etc. Picking a small window size
means we can feed more windows into our model; the downside is that the
model may not have su cient information to detect complex long term
behaviours (if such things exist).
Deep learning models don’t like inputs that vary wildly. Looking at those
columns, some values range between -1 and 1, while others are on the scale
of millions. We need to normalise the data, so that our inputs are somewhat
consistent. Typically, you want values between -1 and 1. The off_high and
volatility columns are fine as they are. For the remaining columns, like
that other blog post, we’ll normalise the inputs to the first value in the
window.
This table represents an example of our LSTM model input (we’ll actually
have hundreds of similar tables). We’ve normalised some columns so that
their values are equal to 0 in the first time point, so we’re aiming to predict
changes in price relative to this timepoint. We’re now ready to build the
LSTM model. This is actually quite straightforward with Keras, you simply
stack componenets on top of each other (better explained here).
model.add(Dense(units=output_size))
model.add(Activation(activ_func))
model.compile(loss=loss, optimizer=optimizer)
return model
Epoch 50/50
6s - loss: 0.0625
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 7/11
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
Caveats aside about the misleading nature of single point predictions, our
LSTM model seems to have performed well on the unseen test set. The
most obvious flaw is that it fails to detect the inevitable downturn when the
eth price suddenly shoots up (e.g mid-June and October). In fact, this is a
persistent failure; it’s just more apparent at these spikes. The predicted
price regularly seems equivalent to the actual price just shifted one day
later (e.g. the drop in mid-July). Furthermore, the model seems to be
systemically overestimating the future value of Ether (join the club, right?),
as the predicted line near always runs higher than the actual line. I suspect
this is because the training data represents a period during which the price
of Ether rose astronomically, so it expects that trend to continue (don’t we
all). We can also build a similar LSTM model for Bitcoin- test set
predictions are plotted below (see Jupyter notebook for full code).
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 8/11
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
The good news is that AR models are commonly employed in time series
tasks (e.g. stock market prices), so the LSTM model appears to have landed
on a sensible solution. The bad news is that it’s a waste of the LSTM
capabilities, we could have a built a much simpler AR model in much less
time and probably achieved similar results (though the title of this post
would have been much less clickbaity). More complex does not
automatically equal more accurate.
The predictions are visibly less impressive than their single point
counterparts. Nevertheless, I’m pleased that the model returned somewhat
nuanced behaviours (e.g. the second line on the eth graph); it didn’t simply
forecast prices to move uniformly in one direction. So there are some
grounds for optimism.
Moving back to the single point predictions, our deep machine artificial
neural model looks okay, but so did that boring random walk model. Like
the random walk model, LSTM models can be sensitive to the choice of
random seed (the model weights are initially randomly assigned). So, if we
want to compare the two models, we’ll run each one multiple (say, 25)
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 9/11
times to get an estimate for the model error. The error will be calculated as
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
the absolute di erence between the actual and predicted closing prices
changes in the test set.
Maybe AI is worth the hype after all! Those graphs show the error on the
test set after 25 di erent initialisations of each model. The LSTM model
returns an average error of about 0.04 and 0.05 on the bitcoin and eth
prices, respectively, crushing the corresponding random walk models.
C rypto C rypto
P rice > P rice ∀ Crypto ≠ OmiseGo,
t t−1
C rypto
where P rice → moon, as t → ∞
t
* This blog does not constitute financial advice and should not be taken as
such. While cryptocurrency investments will definitely go up in value
forever, they may also go down.
Summary
We’ve collected some crypto data and fed it into a supercool deeply
intelligent machine learning LSTM model. Unfortunately, its predictions
were not that di erent from just spitting out the previous value. How can
we make the model learn more sophisticated behaviours?
Change Loss Function: MAE doesn’t really encourage risk taking. For
example, under mean squared error (MSE), the LSTM model would be
forced to place more importance on detecting spikes/troughs. More
bespoke trading focused loss functions could also move the model
towards less conservative behaviours.
Get more and/or better data: If past prices alone are su cient to
decently forecast future prices, we need to include other features that
provide comparable predictive power. That way, the LSTM model
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 10/11
5/10/2018 Predicting Cryptocurrency Prices With Deep Learning - dashee87.github.io
wouldn’t be so reliant on past prices, potentially unlocking more
complex behaviours. This is probably the best and hardest solution.
If that’s the positive spin, then the negative reality is that it’s entirely
possible that there is no detectable pattern to changes in crypto prices; that
no model (however deep) can separate the signal from the noise (similar to
the merits of using deep learning to predict earthquakes). And any pattern
that does appear can disappear as quickly (see e cient market hypothesis).
Just think how di erent Bitcoin in 2016 is to craze-riding Bitcoin of late
2017. Any model built on 2016 data would surely struggle to replicate these
unprecedented movements. All of this suggests you might as well save
yourself some time and stick to autoregression (unless you’re writing a
blog, of course).
But I’m sure they’ll eventually find some use cases for deep learning. In the
meantime, you can build your own LSTM model by downloading the Python
code here. Thanks for reading!
https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/ 11/11