Log In

Bitcoin Prediction



Lets predict the price of Bitcoin using social media data from both Reddit and Twitter with the power of machine learning! We’re going to use a technique called sentiment analysis to find the emotions behind the users and try to see if the price correlates with how people are feeling about the popular cryptocurrency. We’ll use the CoinDesk API to retrieve bitcoin price data and python to program our app.

Code for this video:
https://github.com/llSourcell/bitcoin_prediction

Please Subscribe! And like. And comment. That’s what keeps me going.

Want more education? Connect with me here:
Twitter: https://twitter.com/sirajraval
Facebook: https://www.facebook.com/sirajology
instagram: https://www.instagram.com/sirajraval

More learning resources:







Join us in the Wizards Slack channel:
http://wizards.herokuapp.com/

And please support me on Patreon:
https://www.patreon.com/user?u=3191693

source

40 comments

  1. Swaaagquan 13 February, 2018 at 23:36 Reply

    Lol. You're not a hedge fund, using historic data shows how little use it has for bitcoin. It's a rough measure which is overhyped as a good tool. Most cryptos requires bigger fools.

  2. sidx45 13 February, 2018 at 23:36 Reply

    So I could have social media bots influence sentiment to manipulate perceived speculation. Does this include a bot tracker? Would a bot tracker even work? Would it even matter? If the faked influence really affected the price, the gains would still be real…right?

    Im over thinking this aren't I?

  3. István Nagy 13 February, 2018 at 23:36 Reply

    It is a really great project! I tried out with your price and sentiment data then i collected a couple hundred from the web and i tried with those ones also. Unfortunately LSTM predicts one step and i realized the exchange rate is always one step ahead so this model only "follows" the real price. So when the model predicts a value the exchange rate is already there.
    But I learnt a lot about LSTM, plot, datacollecting and data processing. Thanks a lot!

  4. Angelo Stevens 13 February, 2018 at 23:36 Reply

    Bitcoin 101: Bitcoin has no utility value outside of illegal uses. It's bought on pure speculation, hype, and faith. Bitcoin is a terrible, terrible currency and more merchants accepting it won't change that. What makes a currency good, in order of importance are: stable value, free or virtually free to transact with, wide acceptance.

    You won't get to wide acceptance as long as there is price volatility and while high transaction costs are a big hurdle, it's nothing compared to the volatility issue. And, no, Lightning is not some revolutionary fix, it's a wonky patch that simply introduces new middle men with new fees although those fees should be significantly lower than now, it remains to be seen if they'll be lower than credit/debit card transaction fees.

    But again, the fees, transaction speed and in direct relation, the adoption of BT is all irrelevant if it doesn't have a very stable value, as in as stable or more stable than the USD.

  5. Kazik24X 13 February, 2018 at 23:36 Reply

    Isn't convolutional network better here than lstm? In this approach you would probably use fixed size time window to represent input data anyway and as far as I know CNN is better at recognizing patterns in smaller regions, which is good cause prices depend on short time steps. Also it requires less computation power.

  6. Žiga Franko 13 February, 2018 at 23:36 Reply

    Hi Siraj, really nice to see someone exploring so much with Deep learning algorithms, great work early!
    Also feel free to check my project MIT Black-jack card counting system: https://gitlab.com/franko.ziga/Black-Jack-RL. It uses Deep reinforcement learning to learn how to bet and play in Black-jack game. Currently, its learning to bet as low as possible, but I hope it will soon start making money.
    Thanks for ideas and keep on doing what you do!

  7. SunTrib Fashion 13 February, 2018 at 23:36 Reply

    All the big cryptos will fall like Bitcoin (BTC) and small cryptos will raise. To buy the small cryptos are less risky , less stress and more possiblities to be rich someday. All the big cryptos are corrupt already…

    All the new commers, please buy more cryptos with less money….

  8. Hitesh Vaidya 13 February, 2018 at 23:36 Reply

    @sirajraval the implementation is not clear from github readme. The code you are running is not there in the repo. Could you please give proper directions for implementing this project?

  9. Peter Gormsen 13 February, 2018 at 23:36 Reply

    Hey Siraj, thanks for all your videos! I am studying about deep learning and your videos have been a great help. My question is whether you have tried to test the model on 'test data' – meaning data that is not included for training?

Leave a reply