Congratulations on this, seriously

CodeBuns
CodeBuns

congratulations on this, seriously

All urls found in this thread:

warosu.org/biz/thread/S7843267
warosu.org/biz/thread/S7843267#p7846901
jakob-aungiers.com/articles/a/Multidimensional-LSTM-Networks-to-Predict-Bitcoin-Price

Stark_Naked
Stark_Naked

machine learning

WebTool
WebTool

warosu.org/biz/thread/S7843267

lostmypassword
lostmypassword

warosu.org/biz/thread/S7843267
so do I get this right this is just an LSTM model trained on close prices for 300 epochs?

CouchChiller
CouchChiller

hey man good to see you back. Do you have a blog or a tweeter I can follow? I don't like hoping to randomly run into people on Veeky Forums

Booteefool
Booteefool

Would be useful if you could show this on an actual graph with prices and shit you know

Sharpcharm
Sharpcharm

just

but anyway yes.

b-b-but it's overfitting! a-a-and you can't predict the market!!!
maybe you're right, but my first post on 20th predicted the reversal, and the predictions made on 21, 22, and this one are all consistent with each other.
also the bottom is on 25-26, and on 26th there is ue crypto hearing or something, so it might rebounce there like it did after the sec hearing

no, but i do like shitposting on Veeky Forums

maybe oracuck knows the future, but i surely don't. as long as it predict the major trend, it's enough to make money

Lord_Tryzalot
Lord_Tryzalot

i still think it won't actually touch 6k, but 8k is 99% certain

Boy_vs_Girl
Boy_vs_Girl

no, but i do like shitposting on Veeky Forums

even if you posted them with a 1 week delay (or whatever) it would be cool to follow the progress

Fuzzy_Logic
Fuzzy_Logic

Fair enough, so we going down from here?
Can you show us how this is calculated? Like some console full of numbers or some nerdy shit like that

Emberburn
Emberburn

Its a deep learning neural network. To call this nerdy is the understatement of the decade. This shit is the borg from star trek but it don't need no humans.

VisualMaster
VisualMaster

btw, this is my yesterday position warosu.org/biz/thread/S7843267#p7846901
i've made some comfy games yesterday. i would be at 0.285 if i didn't decide to flip eos, but oh well

if the predictions are completely off i'll just post once a week so you can laugh at me, but if they are right i will probably shitpost and seek attention every day. anyway, i'm just experimenting with money i can afford to lose, so whatever. it's kinda fun

the only code it outputs is the epoch count while training, so nothing interesting to see there

DeathDog
DeathDog

Yeah yeah I just want to see proof that he's indeed doing that and not just drawing random lines

Dreamworx
Dreamworx

Inconclusive proof would be if he posted future prices for us. like say for the next hour.

Lunatick
Lunatick

just
lol, what else would you call 10 lines of R (python, matlab, whathave you).
Anyhow, you built an identity model user, congrats. Google "naive predictor", etc. Essentially your model "learned" a simplified ornstein-uhlenbeck process or the relationship Pt = Pt-1 + e. Where e is the stochastic component.
And sure a naive predictor is right every now and then.

For a more telling test I suggest you do mulit-lookahead predictions throughout your backtest. Lets say every 10 hours (if you are on that frequency) make a 4 hour or longer prediction and see if the direction and price match up with historic data.

The market is absolutely predictable btw, its just not that easy. Otherwise every kid out there with access to keras would be very well off.

TalkBomber
TalkBomber

You are actually retarded, and MLanon is about to make a lot of us a lot of money. Be smart and pay attention. Welcome back. I'm sure you have some great insights about the ongoing dip.

eGremlin
eGremlin

Bit finex just dumped hard.
Did you predict that?

bitfinex 212,071 a few seconds ago 9.75
bitfinex 49,246 a minute ago 9.76
bitfinex 84,565 a minute ago 9.75

HOLY GUAPAMOLE

Sir_Gallonhead
Sir_Gallonhead

don't trust random strangers on internet with your money. i wouldn't.

predicting the next hour is literally impossible imo. i even doubt that big boys like jp morgan can predict actualy close prices. the best thing you can get is the most probable trend for the next couple of days, and that isn't even guaranteed

Essentially your model "learned" a simplified ornstein-uhlenbeck process or the relationship Pt = Pt-1 + e
lol, no, it didn't, stop pulling stuff out of your ass mr smartypants

also lol, i shorted right after running the simulation with yesterday's close, and looks like some whale saw my prediction, kek

farquit
farquit

You are wrong.
There is only 5% chancer to go under 10k

ZeroReborn
ZeroReborn

How the fuck will ML predict massive splurts in volume like this?

I doubt even google could predict that shit

Supergrass
Supergrass

Following closely ML user. This is something I’ve been meaning to get too but I’ve been distracted by working on low latency which is stupid because it’s too competitive but it’s really fun.

You got a repo for this?

Crazy_Nice
Crazy_Nice

lol, no, it didn't
pretty sure it did. What else do you expect it to learn exactly? Some hidden relationship by running a bidirectional lstm lololol? please.

RavySnake
RavySnake

it can't, it's just lucky timing.

and btw, i opened the short 30 minutes ago because i got an entry signal from the accelerator oscillator, which works really good during strong trends, not because the prediction. i use the prediction only to get the trend direction and try catching the top/bottom

PurpleCharger
PurpleCharger

what kind of timeline am I looking at with your chart?

Illusionz
Illusionz

does it help you identify the bottom or do you have to do that yourself?

LuckyDusty
LuckyDusty

idiots overfits on test set
luk guise perfect correlation!
post code or kys

Soft_member
Soft_member

1D

well, i can count the days to the top bottom. my first prediction got right the top at 12.5k more or less, and now by looking and the latest 3, the bottom should be in 2/3 day.

remember that prediction prices is actually a meme, and i doubt it'll go to 6k again.

what i know is that there will be the ue crypto hearing on 26, and that we are in downtrend. i'm not trying to pick the bottom, i'll enter long once i'm sure the trend is reversed by using ta and memelines.

but for today and probably tomorrow i'll short on each rebounce all the way down

also
strong support at 9665.40

PurpleCharger
PurpleCharger

How'd you do it, user? Resources so that I can attempt this myself (books, repos, etc)? I'm a software dev but I have no machine learning experience.

GoogleCat
GoogleCat

Bottom is at 0.5 fib. Literally takes 5 seconds to find out.

Flameblow
Flameblow

learn deep learning. it's a lot of stuff so choose an argument and follow tutorials until you understand how everything works

0.5 is 9k, even without the prediction i'm pretty confident that it will test 8k at the very least (which is 0.658 or something)

RumChicken
RumChicken

Let us know when the bottom is close, we can all make it together.

Gigastrength
Gigastrength

probably right before the ue crypto hearing, or maybe at some specific price like 8k or 9k, or lower. just don't go long until you're sure about the reversal and you'll make it

anyway, i'm going to sleep.

i'll shitpost some more tomorrow if the price will keep dumping, lmao

Sharpcharm
Sharpcharm

for anybody interested
jakob-aungiers.com/articles/a/Multidimensional-LSTM-Networks-to-Predict-Bitcoin-Price
20 lines of code or so minus all the boilerplate. Fortunately enough blogger is not a retard (same can't be said about OP) and actually understands why hes just building a glorified identity model.

Bidwell
Bidwell

I'm gonna be rich

Garbage Can Lid
Garbage Can Lid

So we should expect a retest of 6k? Is there an approx day it's predicting it'll happen?
Thanks for the analysis btw, love your threads

LuckyDusty
LuckyDusty

Because Crypto is so volatile and everything mostly follows bitcoin, would it not be more reliable to predict altcoins reactions to bitcoins prices with ML?

Methshot
Methshot

So it's going below 6k?

Firespawn
Firespawn

bragging about overfitting a few days of training data using the basics of a neural net framework

Fuzzy_Logic
Fuzzy_Logic

not using a deep recurrent Q network

Holy fucking brainlet

BunnyJinx
BunnyJinx

Thoughts on Martin Armstrong?

Fuzzy_Logic
Fuzzy_Logic

His arm is strong.

Disable AdBlock to view this page

Disable AdBlock to view this page

Confirm your age

This website may contain content of an adult nature. If you are under the age of 18, if such content offends you or if it is illegal to view such content in your community, please EXIT.

Enter Exit

About Privacy

We use cookies to personalize content and ads, to provide social media features and to analyze our traffic. We also share information about your use of our site with our advertising and analytics partners.

Accept Exit