Updated chart with yesterday's close and more training epochs

updated chart with yesterday's close and more training epochs

So green was calculted befor blue happened?

green was calculated without knowing the blue line

the absolute state of 6kers

What parameters are you using?

This looks like a LARP chart later fabricated.

Show proof.

...

proof of what?

it's a lstm network after 300 epochs of training

i'll try to find some hourly data so we can test it through the day and see

i'm training it on hourly data, let's see how good/bad it will nail the next 14 hours

thank you for your efforts user, looking forward to seeing more posts

i will if the prediction nails atleast the next 7 days trend.

btw, posting hourly forecast in a moment

do you have a closer view by chance? A 1, 2, or 4 hour chart?

Ah perfect tyy

this is what i get on the hourly chart

predictions are way cooler if you are risking real money, kek

Well I'll be

already creating a recursive mirror of that chart

Pic related
OP

>1m
kek

i'll close my long in exactly 6 hours from now and short there, we'll see how it'll go

something is wrong with your computer screen there's a bunch of lines on it

actually the training data was way longer, like 1 month, including the dip to 6k

LOL
BLOCKLOOP

>What is overfitting?

nah you just panic bought at 11,750 faggot stfu

those are just support/resistance lines lol. need to clear it soon

Whats the difference between original and training data?

it's pretty generous with the closing prices, doesn't look like it's overfitting that much. as long as it nails the uptrend/downtrend it's enough to make money

original is the real btc price i gave in input. then the network trained itself by using the data overlapping the orange line.
after that i asked to predict the green one without knowing the underlining blue one and you see the results (though it used the blue one at the end to verify how accurate the prediction was).
red line is completely made up from the latest data

im guessing it is his model performance on training data.
So basically past performance on data that is included in the model's memory.

exactly, i think i still need to tweak some variables but overall it looks good enough

Is this a program or a manually configured/coded dataset API type deal?

You dipshit.

You need Train/Test/Validation if you want to put a model like this in production, otherwise you are just manually overfitting on your test set.

python and tensorflow, but the model itself is manually coded

i did train/test/validate, what's your point?

also what production? if it works i'll use it to make monopoly money and that's it

If this is legit he wouldn't have sold at 6k and sold at the top like a normal person

Have you tried convolution nets?

kek

What you probably did is Train/Test-split

You've basically overfitted on your test-set by using it for parameter tuning.

The problem with this prediction and most predictions in general is that if an influential amount of people from the outside act upon them, the outcome of the prediction changes dramatically

The market is like a bunch of vectors in a pot, and outside forces just keep throwing vectors in.

Have another? I was a cunthair away from having my short liquidated and now this is happening. We gonna keep dumping?

Well, yeah, theres a bear flag forming (1hr) and we hit resistance + below the cloud on the daily (perfectly) ...dno if your shit is legit or not, but we're going lower for sure when the daily MACD flips red

I think it will bottom out in the 9000s personally

Are you the same user that was posting a few weeks(months?) ago as AI user?

At least you weren't this poor sap 3 hours ago

God damn

i'm doing it as a sideproject, a 1-d convolution net, with convolution layers being like "timeframes" but it's harder to setup properly so it'll take some time

no

i didn't sell at 6k but i've finished the network yesterday

i've updated it with the last hour and it's still more or less the same, though i wouldn't rely that much on the hourly data

There's no way in hell it'll go below 9k. That line of support is simply far too strong. It'll probably go sideways for a good few weeks once it dips below 10k.

so epic user. gonna watch your threads more
what 'language' or platform do you write this in

linuxfag here

the best thing is you can just be a very elaborate troll

so...how low are we going here? 4k? by first week of March?

my thoughts as well lmao

4k is very, very far out of sight at this point. more like 5k absolute bottom. Highly likely bullish continuation imo

It won't even break 10k before 27k in May

stfu moron

python and tensorflow

i actually agree we won't see 4k.

actually i doubt we'll even see 6k. the actual predicted prices are pretty wonky, but 99% it'll drop sub 8k during the next 7 days

Update

It's been more than an hour. Let's see if it was correct.

it's still similar. peak in 4-5 hours and then dump

Post the updated picture.

yeah post up OP

whats it trying to predict? one day in advanced?

for the record, the heigh of the red line in hourly predictions is kinda arbitrary, because my chart was set up for daily values and i can't be bothered fixing it right now

here's the updated one, look like we'll see 11150+ one last time before going down

2 week, but i expect it only to get the main trend of the next week desu

These posts are 100% fud. Trying to get weak hands to short and get ass cleaned. If you think Btc is going back sub $10k before June ish then you’re retarded and high

but i'm long now...

You all realise this is a larp because this idiot sold at 6k right, why on earth would someone make a machine learning program and post it on this shitty site

The absolute state of forever bulls.

You really shouldn't have all inned with your rent money. Americans are going to sell off as soon as they wake.

Bear & bull banter on here is adorable, you guys should just fuck

Topkek go back to being cucked eurofag

How are you importing the data for training? Are you just using price as features?

It would be interesting to test different algorithms with additional features like RSI and other metrics traders use.

Btc in downtrend for months.
Uptrend for a couple weeks

AHAHAHA U MEAN IT MIGHT KEEP GOING UP ? FOREVER BULL UNCNA MAKE MONEY WHEN IT DIPS YOU KNOW ITS ABOUT TOME IT CRASHES AND U KNOW IT.


You sold at 6k and it shows, made more money shorting btc than going, I want it to break support because then I can put a massive short with high leverage and kill it in less than 24 hours, going long is riskier most of the time, but I’m not as deluded as you are weak hands faggot.
Anyone calling for btc crash when it’s above 2 important supports and just had tiny pullback from a long bull run is either coping with the fact he sold low and fudding or straight up brainlet.

Did you miss the -1100 usd drop earlier over the span of 100 minutes or are you just braindead?

Absolutely btfo

yeah, only closes. i grab csv from yahoo finance or coindesk

>It would be interesting to test different algorithms with additional features like RSI and other metrics traders use.
while predicting the real price is just a meme, people predicted moving averages since the 80s
. you can do a lot really

Just bot selling after EMA cross which hits some stop-losses and margin longs.
Now we go up again.

Don't let these fags FUD you OP. I'm sure some hardcore investment firms have been using this tech with crypto and stocks to great effect.

wait is that $1.6 mil or 1.6 mil BTC? brainlet here

Sub 11k again

$1.6 mil

Monitoring this thread, thanks based op

I for one support your endeavors OP these dumb fucks are pretty pathetic trolls

USD lol

thats pretty neat, im trying out a CNN to convolute the time spans instead of features.
whats your ROI?

>whats your ROI?
we'll see it in a week, lol

actually i think that predicting prices is a meme. my next project consists in using deep q learning to train an agent to trade. that's actually doable and provably working

good job user, what kind of network are you building for this? do you have it in git?

>comparing investment firm ML models using advanced stochastics and tuned indicators with OP's shitty lstm/CNN on BTC price.

Here is a tip,basic regression techniques are more effective than any machine learning technique when your dimensionality is low.

How about the OP trains his model up to January with random sampled test data and shows us a blind February prediction.

The salt is delicious.

>going long is riskier most of the time
the absolute state of Veeky Forums

>provably working
have you tried it? i agree some sort of reinforcement learning needs to be tacked on, but for me i cant find the right NN structure thats workable with crypto or stock data. LSTMs didnt work for me, hence CNN. maybe a combo of both.

for provably working i meant that since there exist traders who can consistently make money, then you can train an agent doing the same or even better. i also know some people who did similar stuff for their thesis and now they work as analysts in jp morgan or some banks

What line of education would this be considered?

computer science/financial analyst or something like that

How old are you? When did you start studying this stuff? Just curious and well done user. keep up the good work

Holding onto my short still.. hoping for tendies and not a liquidation email. Fingers crossed!

>Tiny pullback
>1000$ crash in 3hr
>Needed 2 days to get up from 10600 to 11700

This is how you know we are in bear market.

I think you've made the better decision thus far, rising wedge pattern atm

I opened a long at 10800, fuck now I want to close it and go short instead... Hold me Veeky Forums

updated with the latest hour

the red line should be high, it's a charting problem, for actual targets use ta.

it's say it's still consistent enough, predicting another small rally and then dumping.

i'm still keeping my long open for another 3-4 hours and then i'll short, but it doesn't look like it won't touch 11500

almost 24, i'm getting a master degree in cs next year.

i actually started studying programming in 2010 because i wanted to make a minecraft clone, lel

So far honoring both the rising wedge and bearish consolidation/pennant.. leaning toward shorting atm myself

user. Can you create a discord and share your based technicals with us? You seem to have a good handle on this tool.

Top and bottom diag lines are the rising wedge lines

That red is below the earlier lows? I shorted that spectacularly and then got greedy. Paid for it with a sleepless night and a 9-5 I gotta be at shortly. Yuck.

the red line should be higher and lefter, it's a charting problem

So many lines can barely tell what's what