First AI to beat multiplayer Poker game vs top pros (like chris furgeson, darren elias...)
09-12-2019
, 06:01 AM
Quote:
Trust the machines and HODL
https://medium.com/datadriveninvestor/predicting-cryptocurrency-prices-with-machine-learning-1b5a711d3937
https://medium.com/datadriveninvestor/predicting-cryptocurrency-prices-with-machine-learning-1b5a711d3937
Juk
09-12-2019
, 09:35 AM
Could you elaborate?
09-12-2019
, 04:09 PM
1. He says he has a "big dataset" containing "60 thousand different data points" when actually he has tiny dataset containing 2667 data points (days) with 37 features per day...
2. He then forgoes trying any simple models and jumps right in there with a "Deep Learning Model" with millions (billions?) of tunable parameters:
3. He then claims his model has a high directional accuracy, without any reference to a baseline model of "always predict up", etc (and just looking at the graph it's just been trending upwards the whole way anyway lol...).
4. He then proceeds to show a graph that looks like the predicted price has no real correlation with the actual price, but just skips over this troubling revelation without actually bothering to measure the correlation anyway...
5. He then follows this up with another graph purporting to show how well his model follows the price; when in actual fact it looks like it's mostly just predicting "tomorrows price = today's price + some random noise"...
6. Then finally concludes:
In general just avoid anything on medium.com (and towardsdatascience.com), as the articles are either just total trash like this or copy-and-pasted beginner tutorials put up by people hoping to add something to their resume.
But, if you are actually interested in forecasting then there is a really good (free) book available here: Forecasting: Principles and Practice.
Juk
2. He then forgoes trying any simple models and jumps right in there with a "Deep Learning Model" with millions (billions?) of tunable parameters:
Quote:
Input layer (takes data of shape n samples x 50 x 37)
Bidirectional LSTM layer (returns a sequence, 100 cells)
Dropout layer (20% dropout — reduces overfitting)
Bidirectional LSTM layer (returns a sequence, 100 cells)
Dropout layer (20% dropout — reduces overfitting)
Bidirectional LSTM layer (doesn’t return a sequence, 50 cells)
Output layer (returns the predicted next day price of Bitcoin)
Bidirectional LSTM layer (returns a sequence, 100 cells)
Dropout layer (20% dropout — reduces overfitting)
Bidirectional LSTM layer (returns a sequence, 100 cells)
Dropout layer (20% dropout — reduces overfitting)
Bidirectional LSTM layer (doesn’t return a sequence, 50 cells)
Output layer (returns the predicted next day price of Bitcoin)
4. He then proceeds to show a graph that looks like the predicted price has no real correlation with the actual price, but just skips over this troubling revelation without actually bothering to measure the correlation anyway...
5. He then follows this up with another graph purporting to show how well his model follows the price; when in actual fact it looks like it's mostly just predicting "tomorrows price = today's price + some random noise"...
6. Then finally concludes:
Quote:
What does this mean?
I created a machine learning model that was able to predict cryptocurrency models with reasonably high accuracy by finding intricate patterns in highly complex data. This shows how powerful machine learning is and how it has a huge variety of applications. Using this model itself on cryptocurrency would allow people to make a lot of profits by letting them buying and selling cryptocurrencies at predicted times.
I created a machine learning model that was able to predict cryptocurrency models with reasonably high accuracy by finding intricate patterns in highly complex data. This shows how powerful machine learning is and how it has a huge variety of applications. Using this model itself on cryptocurrency would allow people to make a lot of profits by letting them buying and selling cryptocurrencies at predicted times.
But, if you are actually interested in forecasting then there is a really good (free) book available here: Forecasting: Principles and Practice.
Juk
09-12-2019
, 04:22 PM
Interesting article I found today relevant to this thread:
Superhuman AI Bots Pose a Threat to Online Poker Firms, Morgan Stanley Says
Juk
Superhuman AI Bots Pose a Threat to Online Poker Firms, Morgan Stanley Says
Quote:
While online gaming firms like The Stars Group, GVC Holdings Plc and Playtech Plc have safeguards in place to protect against bots, a loss of player confidence and ultimately revenue from a proliferation of AI bots are “plausible risks,” Young said. The broker trimmed price targets for all three stocks, with Stars the most exposed to online poker.
03-06-2023
, 08:25 PM
03-07-2023
, 01:48 AM
03-07-2023
, 03:53 PM
Quote:
Interesting article I found today relevant to this thread:
Superhuman AI Bots Pose a Threat to Online Poker Firms, Morgan Stanley Says
Juk
Superhuman AI Bots Pose a Threat to Online Poker Firms, Morgan Stanley Says
Juk
Wtf is a "superhuman" a.i bot, anyway ?