I have an algorithm that can predict peaks and dips with 99.75% accuracy

it repaints, oops!

can it predict chink scams tho, and pajeet shit shillers?

It only works for bitcoin because that's the only coin with enough historical data.

Can you help out an user who is in bad debt make some btc?

Attached: cool story.png (376x392, 208K)

Acknowledged.

I'll take it

Attached: 1532284839278.jpg (639x446, 17K)

You just overfitted on historical data. Backtesting is a meme, and actually hard to do right.

But it generalizes outside my training set. I agree that the accuracy seems suspiciously high, but so far everything seems correct.

Did you use NLP on social media, ML or something else?

How’s your test set? You must’ve gotten 99%+ on training set, right? Did you overfit user?

is it repainting on a forward test? (tickdata)

Not using NLP or neural networks. It is ML though.

What exactly do you mean by repainting?

post pics i dont believe you

>outside muh training set

Major cryptos are all correlated. They’re all the same graph

Jow Forums pls go.

Attached: 1528495065382.jpg (723x800, 93K)

what indicators do you use?

It generalizes to a different part of the same bitcoin time series.

They're not publicly accessible. I'm doing a ton of preprocessing on the data before I feed it into my model.

> muh ML
> generalizes outside of my training set
> 99% accuracy
> everything seems correct
just imagine being this fucking retarded... god

RSI.

There probably is something wrong with it, but we'll see.

> I'm doing a ton of preprocessing on the data before I feed it into my model.
oh there it is
don't tell me you do standardization/scaling of timeseries data before processing it with the model

There's a bit more to it than that.

yeah thats not my point
I bet you are introducing lookahead bias without even knowing it

hey bro

all the luck to you

how soon will you guarantee 100% ?

so I can try it out

I don't see how I could have with the way I implemented it. I can give it another look though.

I probably won't be releasing it if it does indeed work like it seems to.

so if your not going to share you are here to gloat ?

>nice

No, I just think there should be more interesting discussions on Jow Forums.

checked

Attached: 1517970326251.png (733x464, 131K)

since youve mentioned ML you are mostly working with the mean of the distribution, as in you are predicting the mean (assuming regular cost functions, squared loss, mse, hinge, etc). Now, unless some rando on Jow Forums made a groundbreaking discovery, in which case you are welcome to submit your algo to M competitions and similar, the GBM still applies. Which should tell you a thing or two. Money is made predicting the tails

unless you are talking about HFT and similar. There the process generally moves away from GBM/exponential kernel and takes on another kernel (powerlaw, etc). I can do 80% accuracy predicting 30sec ahead with xgboost and order book flow. But good luck bridging the spread and fees.

Well what the fuck is it? Either tell us or leave.

Yeah we'll see. It definitely needs more testing. The special sauce isn't in the ML though: I just got tired of hand tuning everything.

thanks for sharing

to generalize

where is btc going ?

should we still expect a dump or are we green from here ?

ty

Attached: wth.jpg (560x352, 63K)

lol alright man, good luck with your special sauce, thanks for the "interesting discussion" I guess

Wtf you guys use to build this? Python? How do I get started?

Attached: Offered-up-for-sex.jpg (600x899, 71K)

charts.woobull.com/bitcoin-nvt-signal/

you're welcome Jow Forums

Attached: 14.jpg (960x545, 101K)

it doesnt matter, build it in cobol if you like. What matters is the underlying model

>predict peaks and dips
>still buying high and selling low

Attached: images.png (211x239, 6K)

You guys are pathetic

Attached: pepe.jpg (227x222, 6K)

tradeview is full of strategies with >95% accuracy but they don't work on live data
you should test it, thats what we call repaint

I, as a creator of a BTC peak/dip predictor too, am going to say that what you are reporting could be true and may be entirely possible, kek

e.g. my algorithm has been working since early 2016, and so far, as of now, +60,000% in terms of BTC

by the way, it's magical that great minds think alike, kek

Predict outcomes you say? I wonder. What the chances are that my team wins the dark tournament.

Attached: unnamed (4).gif (182x114, 15K)

noticed

How the fuck do you predict this?

Attached: predict.png (777x548, 20K)

99%, care about the masked dude

right? It took me 2 years to finally come up with a long term (80% accuracy, but cuts losses, and lets gains run) indicator that doesn't repaint from higher time periods that have already happened. uses pine v3 open[1] or close[1],lookahead=barmerge.lookahead_off

Attached: norepaintingallowed.png (361x56, 3K)

If its so accurate run it out 3 years and tell us all the prices and what dates they hit on.

Otherwise nice larp and fuck off.

ty

>OP says he was right 99.75% of the time
>nobody asks him about the other 0.25%
Seems pretty fucking important..

post pine

market make or get left behind.

im profitable since last october with the same strategy and made it with an investment of 5k, which was the fee i paid to get to a decent volume to get a reduction in fees.

gl

Hand it over. That's not for your kind.

Attached: robobog.jpg (540x800, 293K)

Mine’s only around 50% it almost fucked me 3 months ago. For the noobs, its written in python