/sci/ - Science & Math » Thread #12089126

15KiB, 387x248, download.png

View Same Google iqdb SauceNAO

Anonymous Sat 05 Sep 20:58:36 2020 No.12089126 View Reply Original Report

Quoted By: >>12089144 >>12089148 >>12089212 >>12089243 >>12089320 >>12089330 >>12091029 >>12092164 >>12092210 >>12092876 >>12093067

My bitcoin prediction algo. Discuss

Anonymous

Anonymous Sat 05 Sep 2020 21:01:26 No.12089139 Report

Quoted By:

Cool if real. Seems to lag behind the real data though. Or have you shifted the blue line to make it more visible?

Anonymous

Anonymous Sat 05 Sep 2020 21:02:13 No.12089144 Report

Quoted By: >>12089160

>>12089126

So where we gonna go over the next 5 months?

Anonymous

Anonymous Sat 05 Sep 2020 21:03:26 No.12089147 Report

Quoted By: >>12089166 >>12089181

It lags because its an autocorrelation. It predicts a morph of the current price

Anonymous

Anonymous Sat 05 Sep 2020 21:03:44 No.12089148 Report

Quoted By: >>12089154

>>12089126
file name

Anonymous

Anonymous Sat 05 Sep 2020 21:04:51 No.12089154 Report

Quoted By:

>>12089148
Sorry thats what jupyter gives

Anonymous

Anonymous Sat 05 Sep 2020 21:06:17 No.12089160 Report

Quoted By: >>12091071 >>12092867

>>12089144
Idk it predicts a day in advance

Anonymous

Anonymous Sat 05 Sep 2020 21:07:31 No.12089166 Report

Quoted By: >>12089170 >>12089178

>>12089147
So can you extrapolate data from this? Have you tried it? Were your predictions accurate? Or is this just some meme fit of the data?

Anonymous

Anonymous Sat 05 Sep 2020 21:08:17 No.12089170 Report

Quoted By:

>>12089166
It looks like a meme fit, otherwise he wouldn't post this here. He'd be on /biz/ gloating about his crypto fortune.

Anonymous

Anonymous Sat 05 Sep 2020 21:09:14 No.12089178 Report

Quoted By:

>>12089166
Im still tuning hyperparameters, havent done a backtest

Anonymous

Anonymous Sat 05 Sep 2020 21:09:52 No.12089181 Report

Quoted By: >>12089197 >>12092757

>>12089147
>autocorrelation
yeah this is a great idea, dump all your money into it and see how that goes. I'm sure a massive price correction wouldn't affect you in the slightest

Anonymous

Anonymous Sat 05 Sep 2020 21:13:10 No.12089197 Report

Quoted By: >>12089202

>>12089181
You can make like 5% profit a week, price corrections dont happen often

Anonymous

Anonymous Sat 05 Sep 2020 21:14:24 No.12089202 Report

Quoted By:

>>12089197
you're right, go for it anon. remember me when you're the world's first trillionaire.

Anonymous

Anonymous Sat 05 Sep 2020 21:17:55 No.12089212 Report

Quoted By: >>12089217

>>12089126
github link or btfo

Anonymous

Anonymous Sat 05 Sep 2020 21:20:24 No.12089217 Report

Quoted By: >>12089229 >>12091102

>>12089212
Its a simple neural network that uses the price

Anonymous

Anonymous Sat 05 Sep 2020 21:26:17 No.12089229 Report

Quoted By: >>12089238

>>12089217
btfo

Anonymous

Anonymous Sat 05 Sep 2020 21:29:23 No.12089238 Report

Quoted By:

>>12089229
Uwu

Anonymous

Anonymous Sat 05 Sep 2020 21:31:27 No.12089243 Report

Quoted By: >>12089247

>>12089126
is this training data or held-out data?

Anonymous

Anonymous Sat 05 Sep 2020 21:32:32 No.12089247 Report

Quoted By: >>12089270

>>12089243
Test data, not train

Anonymous

Anonymous Sat 05 Sep 2020 21:40:34 No.12089270 Report

Quoted By: >>12089279

>>12089247
Are you sure you are not "cheating", i.e. using any kind of information during training that can only be known during the forecasting horizon? It seems pretty fishy that the prediction follows all hills and valleys. Your prediction is more or less just the true values smoothed and lagged a bit. Either you are lying on the internet or you use "future" information at some point without even realizing.

Anonymous

Anonymous Sat 05 Sep 2020 21:42:01 No.12089273 Report

Quoted By: >>12089286

> looks where price currently is.
> adjusts prediction to price

Anonymous

Anonymous Sat 05 Sep 2020 21:43:24 No.12089279 Report

Quoted By: >>12089285 >>12092961

>>12089270
No, i didnt. Btw the chart shows a month of the price

Anonymous

Anonymous Sat 05 Sep 2020 21:44:48 No.12089285 Report

Quoted By:

>>12089279
post code nigger

Anonymous

Anonymous Sat 05 Sep 2020 21:44:50 No.12089286 Report

Quoted By:

>>12089273
It looks and adjusts sir

Anonymous

Anonymous Sat 05 Sep 2020 21:56:42 No.12089320 Report

Quoted By:

>>12089126
Why are they not starting at the same point.
I'm calling bullshit.

Anonymous

Anonymous Sat 05 Sep 2020 21:59:50 No.12089330 Report

Quoted By: >>12089345

>>12089126
It is shit because it does not predict trends. The blue line (prediction) is always lagged compared to the red (market data). You are better off using some moving average as your predictor.

Anonymous

Anonymous Sat 05 Sep 2020 22:03:51 No.12089345 Report

Quoted By: >>12089356 >>12092867 >>12092873

>>12089330
It predicts the trends at the high points

Anonymous

Anonymous Sat 05 Sep 2020 22:06:47 No.12089356 Report

Quoted By: >>12089376

>>12089345
I think you should use a median filter before training your neural net.

Anonymous

Anonymous Sat 05 Sep 2020 22:11:34 No.12089376 Report

Quoted By:

>>12089356
It smooths the data fine on its own doesnt it

Anonymous

Anonymous Sat 05 Sep 2020 22:15:18 No.12089384 Report

Quoted By:

>download.png
Nice try fatass

Anonymous

Anonymous Sun 06 Sep 2020 11:29:45 No.12091029 Report

Quoted By: >>12091055

>>12089126
Plot the diff of the two lines.
Realise it's worthless piece of shit.

Anonymous

Anonymous Sun 06 Sep 2020 11:45:40 No.12091055 Report

Quoted By: >>12091063

>>12091029
you don't need that, OP is either a troll or a retard who doesn't realize that btc value is inflated because of hype

it was as high as 19.5k before exactly because there were retards willingly buying it at that price due to hype

Anonymous

Anonymous Sun 06 Sep 2020 11:50:13 No.12091063 Report

Quoted By: >>12091066

>>12091055
Well, I wanted the OP to plot it, as I just developed much better predictor. I'm sure it will have far smaller error average and also smaller standard deviation.
For those keen on investing, you just take current price as your next step price.

Anonymous

Anonymous Sun 06 Sep 2020 11:51:29 No.12091066 Report

Quoted By: >>12091195

>>12091063
your predictor is shit and you're a retard like op, fuck off nigger.

Anonymous

Anonymous Sun 06 Sep 2020 11:54:50 No.12091071 Report

Quoted By:

>>12089160
So feed the output back into the input even though accuracy declines for multiple days forward

Anonymous

Anonymous Sun 06 Sep 2020 12:09:51 No.12091102 Report

Quoted By:

>>12089217
It most likely just memorized the curve, that's exactly the reason why you need to split the data into the training and test data.

Anonymous

Anonymous Sun 06 Sep 2020 12:56:25 No.12091195 Report

Quoted By:

>>12091066
It's less shit than the OP's.
Go make any predictor and invest in bitcoin, good luck.

Anonymous

Anonymous Sun 06 Sep 2020 18:12:27 No.12092164 Report

Quoted By: >>12092867

>>12089126
>price at time t = price at time - 3 days

based

Anonymous

Anonymous Sun 06 Sep 2020 18:29:26 No.12092210 Report

Quoted By:

>>12089126
You """invented""" a moving average.

Anonymous

Anonymous Sun 06 Sep 2020 21:31:12 No.12092757 Report

Quoted By:

>>12089181
Are you stupid

Anonymous

Anonymous Sun 06 Sep 2020 22:15:37 No.12092867 Report

Quoted By: >>12092873

>>12089345
>It predicts the trends at the high points
It's literally as good at prediction as
>>12092164
kek
If I were to simulate this by starting from day 1 and follow the prediction 1 day ahead as you say here
>>12089160
at no point does it predict anything at all useful, I can't see a single point in time where it would be helpful

Anonymous

Anonymous Sun 06 Sep 2020 22:17:17 No.12092873 Report

Quoted By:

>>12092867
>>12089345
Wait I lied, it actually predicts the market will go high when the market goes low one day ahead and vice versa. It's actually worse to follow this predictor than to just flip a coin for up/down and go with your gut feeling

Anonymous

Anonymous Sun 06 Sep 2020 22:17:57 No.12092876 Report

Quoted By:

>>12089126
yea sure anon, seems legit

Anonymous

Anonymous Sun 06 Sep 2020 22:36:56 No.12092961 Report

Quoted By: >>12093127

>>12089279
But it's just a test set, so past values. You can't replicate it when you do a prediction, wait for a month and compare true and predicted. Because you can't even do a prediction into the future today (instead of using test sets). Because when you attempt that, you will realize that your model used some lag values that that are not known in the present. E.g. you don't know the 20-day-lag in 21 days. Moron.

Anonymous

Anonymous Sun 06 Sep 2020 23:20:24 No.12093067 Report

Quoted By:

>>12089126
Looks like baby's first time series analysis

Anonymous

Anonymous Sun 06 Sep 2020 23:38:21 No.12093127 Report

Quoted By:

>>12092961
You use the model by buying if the price tomorrow is higher or selling if the price tomorrow is lower

Anonymous

Anonymous Mon 07 Sep 2020 01:04:10 No.12093325 Report

Quoted By:

Op here I ran a backtest and it made money

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches