# Let’s teach our deep RL agents to make even more money using feature engineering and Bayesian optimization

W e just launched AlgoHive, an open-source project to crowdsource the prediction of cryptocurrency prices and automate crypto trading. We are now sharing our vision towards where our project is headed.

To that end I have laid out a plan that accomplishes the above while providing a lot more structure to our efforts. AlgoHive is a free open-source community and will always be as far as I am concerned. To that end I have been approached by several people in the AI and crypto space that have been impressed by what we are building. Based on this level of interest and our rapid community growth I would like to explore possible ways to ultimately generate revenue from our efforts.

There are several revenue models that may make sense which I will share soon. What I am sharing now is our initial draft of a multi-year vision in what I believe AlgoHive will become. We certainly have all the right people on the bus.

Think of AlgoHive as a crowdsourced tech startup that we are all a part of, learning as we go along and collectively pivoting as needed. Many thanks to the countless hours that our project team has put in to make this launch successful. The projects listed here are all members of Project AlgoHive, a community of data scientists, machine experts, crypto traders, financial analysts, and angel investors that are working together to help make cryptocurrency trading safer, smarter and easier.

AlgoHive hopes to bring these ideas closer to reality not by one person or small private team but with a scalable crowdsourced model that has never been attempted in this capacity. So many others wanted to learn how to be smarter about crypto trading. After many, many conversations with other traders and data experts I discovered that many of us are actually trying to solve the same problem. This led me to build a collaborative platform to bring other great crypto minds together and launch the AlgoHive project and community.

This is free because this is the resource that I would have loved to have years ago but unfortunately I needed to learn it all by myself, make many costly mistakes and had no one to share my ultimate successes with. Although this is mostly a collaborative approach all the expenses of research, web development, web hosting, community management, developing and testing new algorithms, finding data sources I pay for out-of-pocket. While I want to forward the crypto movement the only way that this will be sustainable in the beginning is via donations.

To that end I am creating an AlgoHive public wallet crypto donations which will be publicly viewable by anyone. Anyone will be able to see real-time any funds used.

Ultimately I will align project goals to a crowdfunding initiative for continued project development. Although crypto donations are very welcome, we also highly value community contributions. Below is our growing list of AlgoHive community projects that are at the intersection of crypto and machine learning and with promising uses of alternative data sources:.

For more detail including our project roadmap and latest projects visit our Github page. If you found this article to be helpful please Clap up to 50 times and share to help get it in front of more smart people like yourself. If you would like to get the latest updates on our project please follow me here on Medium.

If you want to be a part of Project AlgoHive, a crowdsourced cryptocurrency prediction startup you can learn more here. Marc Howard marcbegins. Tweet This. Continue the discussion. Marc Howard Oct Marc Howard Nov Hackernoon Newsletter curates great stories by real tech professionals Get solid gold sent to your inbox. Every week! Aruna Gomathi Aug Alex Wang Mar Kenny Li.

Thomas Wiradikusuma Sep Contact Us Privacy Terms.

## Recurrent Networks

In this post, we introduce Keras and discuss some of the major obstacles to using deep learning techniques in trading systems, including a warning about attempting to extract meaningful signals from historical market data. In the last few years, deep learning has gone from being an interesting but impractical academic pursuit to a ubiquitous technology that touches many aspects of our lives on a daily basis — including in the world of trading. This meteoric rise has been fuelled by a perfect storm of:. Deep learning excels at discovering complex and abstract patterns in data and has proven itself on tasks derp have traditionally required the intuitive thinking of the human brain to solve. That is, deep learning is solving problems that have thus tradign proven beyond the ability of machines. Therefore, it is incredibly biycoin to apply deep learning to the problem of forecasting the financial markets.

## Recurrent Networks

GitHub is home to over 40 million developers working together to host and review code, manage projects, and build software together. If nothing happens, download GitHub Desktop and try again. If nothing happens, download Xcode and try again. If nothing happens, download the GitHub extension for Visual Studio and try again. Exploiting Bitcoin prices patterns with Deep Learning. Like OpenAI, we train our models on raw pixel data. Exactly how an experienced human would see the curves and takes an action. Training on 5 minute price data Coinbase USD. Some examples of the training set.

## Why you should be cautious with neural networks for trading

This post brings together cryptos and deep learning in a desperate attempt for Reddit popularity. If you were to pick the three most ridiculous fads ofthey would definitely be fidget spinners are they still cool? But enough about fidget spinners!!! 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!

I thought this was a completely unique concept to combine deep learning and cryptos blog-wise at leastbut in researching this post i. And since Ether is clearly superior to Bitcoin have you not heard of Metropolis?

If you wish to truly understand the underlying bitcin what kind tradingg crypto enthusiast are you? Before we build the model, we need to obtain some data for it. In deep learning, no model can overcome a severe lack of data. Before we import the data, we must load some python packages that deeo make our lives so much easier. With a little bit of data cleaning, we arrive at the above table. Bitcoin trade 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.

You can see that the training period mostly consists of periods when cryptos were relatively cheaper. But why let negative realities get in the way of baseless optimism? Extending this trivial lag model, stock prices are commonly larning as random walkswhich can be defined in these mathematical terms:. 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.

The Bitcoin random walk is particularly deceptive, as the scale of the y-axis is quite wide, making the prediction line appear ldarning smooth.

Single point predictions are unfortunately quite common when evaluating time series models e. A better idea could be to measure its accuracy on multi-point predictions.

Thus, poor models are penalised more heavily. In mathematical terms:. The model predictions are extremely sensitive to the random seed. In the accompanying Jupyter notebookyou can reep play around with the seed value below to see how badly it can perform. TensorFlowKerasPyTorch. The volatility columns are simply the difference between high and low price divided by the opening price. We must decide how many previous days it will have access to. We build little data frames consisting of 10 consecutive days of data called windowsso the first window will consist of the th rows of the training set Python is zero-indexedthe second will be the rows.

Picking a small window size means we can feed more windows into our model; the downside is that the model may not have sufficient information to detect complex long term visit web page if such things exist. 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. This is actually quite straightforward with Keras, you simply stack componenets on top of each other better explained. The function also includes more generic neural network features, like dropout and activation functions.

Trqding start by examining its performance on the training set data before June Instead of relative changes, we can view the model output as daily closing prices.

The model learniing access the source of its error and adjust itself accordingly. We should be more interested in its performance on the test dataset, as this represents completely new data for the model. 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. The predicted price regularly seems equivalent to the actual price just shifted learnijg day later e.

Furthermore, the model seems to be systemically overestimating the future value of Ether join the club, right? We can also build a similar Learnung model for Bitcoin- test set predictions are plotted below see Jupyter notebook for full code. Our fancy deep learning LSTM model has partially reproducted a autregressive AR model of some order pwhere future values are simply the weighted sum of the previous p values. Depe can define an AR learnin in these mathematical terms:. The good news is that AR models are commonly employed in time series tasks bitcoin trading fee. More complex does not automatically equal more accurate.

The predictions are visibly less impressive than their single point counterparts. 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. The error will be calculated as the absolute difference 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 bitcoin trading deep learning 25 different initialisations of each model.

The LSTM model returns an average error of about 0. Aiming to beat random walks is a pretty low bar. While cryptocurrency investments will definitely go up in value forever, they may also go. Unfortunately, its predictions were not that different from just spitting out the previous value. How can we make the model learn more sophisticated behaviours? More bespoke trading focused loss functions could also move the model towards less conservative behaviours.

Easier said than done! This is probably the best and hardest solution. And any pattern that does appear can disappear as quickly see efficient market hypothesis. Just think how different Bitcoin in is to craze-riding Bitcoin of late Any model built on data would surely struggle to replicate these unprecedented movements.

Thanks for reading! This post describes two popular improvements to the standard Poisson model for football predictions, collectively known as the Dixon-Coles model. Announcing my new Python package with a look at the forces involved in cryptocurrency prices. This post investigates the universally known but poorly understood home advantage and how it bigcoin in football leagues around the world.

Taking a break from deep learning, this post explores the recent surge in song collaborations in the pop charts. Predicting Cryptocurrency Prices With Deep Learning This post brings together cryptos and deep learning in a desperate attempt for Reddit popularity.

David Sheehan Data scientist interested in sports, politics and Simpsons references. Follow London via Cork Email Github. Data Before we build the model, we need to obtain some data for it. Leave a Comment. Analysing the Factors that Influence Cryptocurrency Prices with Cryptory 15 minute read Announcing my new Python package with a look at the forces involved in cryptocurrency prices.

Home Advantage in Football Leagues Around the World 10 minute read This post investigates the universally known but poorly understood home advantage and how it varies in football leagues around the world.

Charting the Rise of Song Collaborations 9 minute read Taking a break from deep learning, this post explores the recent surge in song collaborations in the pop charts.

## Bitcoin Trading Bot (Tutorial)

## Long Short Term Memory (LSTM)

To that end Derp have leraning approached by several people in the AI and crypto space that have been impressed by what we are building. Drawdown is the measure of a specific loss in value to a portfolio, from peak to trough. Tweet This. Thanks for reading! Hold on to your seats everyone, this is going to be a wild ride. Become a member. This is free because this is the resource that I would have loved to have years ago but unfortunately I needed to learn it all by myself, make bitfoin costly mistakes and had no one to share my ultimate successes. Now, I am no fool. About Help Legal. Watching this agent trade, it was clear this reward mechanism produces strategies that over-trade and are not capable of capitalizing on market opportunities.