You are viewing a single comment's thread from:
RE: Bitcoin neural network analysis week 20/2019: Short squezzed
Did you mention earlier which are the criteria on how the network is speculating? I have no clue on what this is build on. I'm simply impressed how accurate it is working.
Greets
Posted using Partiko Android
The network is trained on past price data since 2014 and is looking at the prices of the past 100 days to estimate the price of tomorrow.
Internally it is a black-box and I have no idea how it is coming up with the numbers. But here is a useful plot
.png)
These are normalised relative price changes in the training sample in blue and the network prediction in orange. It is immediately clear that they differ a lot and the network does not perfectly predict the price at all. But there is still a clear correlation between the two curves so that the network does predict a small hint.
Okay, I see now what is the black-box and network. How does the training looks like? The network gets a training sample and reacts according to which principal? Is there a scalation?
Posted using Partiko Android
The network is training on past data trying to minimise the post-diction error.
Okay. I guess to see now the intuition in it.
Thanks for the post✌️
Greets
Posted using Partiko Android