Share this page!

Enter Twitter Thread URL to Unroll

Needs to be the full URL like: https://twitter.com/threadreaderapp/status/1644127596119195649

How to get URL link on Twitter App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Paul Smith 🌷🇬🇧 Profile picture
@PJSmith
Jun 15 • 13 tweets • 3 min read Twitter logo Read on Twitter
Work in progress
KNN ML - Stores all the patterns of past bars, then iterates over all those patterns finding the closest X matches to now, and from those matches, calculates the probability of what happens next. Simple ML, but quite well suited to this, I think Image
You do have to believe, however, that past history is in fact reflective of what might happen in the future...
So, I had quite a bit of fun (only smashed 3 things!) with this yesterday. I like it. I think it gives me real-time probabilities, which is what I am always looking for. Anyway - Now I ask for some ideas from you 😁 Image
If you could consider just a few metrics to determine short-term price action, what would they be?
I've tried ROC, Vol, and above/below mean. Other ways might be, abve/blw the open/vwap, in or out of value, etc.
Any suggestions? My eyes have a little more blood left in them yet😂
Got some questions. So A Simple explanation of what this is. Its basically a vote. Ask 100, or 200
or whatever people what will happen based on their past experience, and go with the majority.
Could not help myself. I am not trading but just did a 1 lot. Quick scalp. Testing software! lol. #coffeemoney Image
I promise to take no more trades!
For the techies amongst you. This is how long it takes to make a prediction with around 10,000 samples in the dataset. So, quite capable of running on small time frame bars in real-time, and this is still dev code

Result took : 00:00:00.0040113
Itteration #1,239,437 Image
On a 3 x larger timeframe chart, just to see...
Looks good to me, but I have been fooled before!
No settings or optimisation for this change, btw - It's all full auto! Image
Increased training period to 60 days and window to 10k bars. Looks better (even though it was already good!), and with a few tweaks, is still way fast!

Result took : 00:00:00.0025595

I can save this now, and put the chart period back to just a few days, to save NT from dying!
Sorted! 60 days of training data on a 5-day chart
Sometimes I even impress myself... Image
btw - I will add my own 2 brain farts to this list

Wicks on candles
USD

If you've already tried this, please lmk

• • •

Missing some Tweet in this thread? You can try to force a refresh
　

Keep Current with Paul Smith 🌷🇬🇧

Paul Smith 🌷🇬🇧 Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @PJSmith

Paul Smith 🌷🇬🇧 Profile picture
Jun 16
Figured out how to save and restore model data (arrays mainly) efficiently. So, now, I can share and distribute trained models!
Beer really does work... Image
Trained for 80 range and hooked in with profileBars. 90 days of training data on the 7-day chart. Awesome companion! (zones are from HeikenAshi indie) Image
You know what's coming next, don't you, if all this works out?
Read 4 tweets
Paul Smith 🌷🇬🇧 Profile picture
Jun 11
Was reading some articles on the MQL site yesterday and it got me to thinking.. (btw - even though I do not use the platform, MQL has articles and user contributions the other platforms like Ninjatrader can only dream about!)
So, some research starts
Anyway. Looking at short term trend probabilities over a set number of brick style bars (12 range in this case). This histogram represents the imbalance of up or down closes in that range. I am just going to let that sit for a while, and marinate... 🍸 Image
With a bucket size of 2 Image
Read 12 tweets
Paul Smith 🌷🇬🇧 Profile picture
May 8
Very narrow range. Magnet below. Singles need fixing. If breaks with some follow-through, it's the logical spot (but when has NQ ever been logical!?). Image
Singles are all filled. Took some off at the end. Some magnets are about 10-20pts lower, but...
Scale a little more into vpoc. Not much left now
Read 4 tweets
Paul Smith 🌷🇬🇧 Profile picture
May 7
Came up with a completely new way to score chart /indicator patterns yesterday. Just finished coding it. Works on reversal-type bars. The idea is, you take your pattern, then add or subtract the Abs(Open-Close) of the next bar. This gives you a little score on continuation bars
..and a big score on reversal bars (because, of the reversal-type chart). Iterate through all your bars and each pattern ends up with a sum of the scores. Some patterns will be wildly positive, some negative. You're in the middle is chop. Auto-biased for the current action.
..this should give a statistical probability of if the next bar is up, or down. But, this probably is not to trade-off. It IS to say, if this indie/pattern is saying this, the data really does show, the next bar(s) will be up/down. Data to back up your beliefs. ImageImage
Read 4 tweets
Paul Smith 🌷🇬🇧 Profile picture
Apr 4
#Nasdaq Levels share (I added a legend!)
Clear as mud. I am out today, so hope you just keep on playing tennis until I get back...
Just back. You could have bloody waited!
Read 4 tweets
Paul Smith 🌷🇬🇧 Profile picture
Feb 26
I try all sorts of whacky stuff on weekends. Most of it ends up being totally useless. I waste far too much of my life on this stuff!
Currently backtesting over the last 2 weeks with my day-to-day trend algos. Decided to break my order into parts and spread it over 100 nq pts 1/
Result? A mediocre profit turned into a much larger profit 🤔
The same risk control and stop were used. Only the initial entry size was much reduced. A much higher profit was attained. A sizeable one-way drive over 100pts would probably hurt 2/
But, not I think that much. Still cuts if the trend changes. More testing to do, but throwing the concept out there, into the wind, as I tend to do. Maybe you algo guys are interested. I am sure it's been around for a while, and perhaps someone here has already tried this?3/
Read 5 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!

Send Email!

:(