I've been looking for a channel like this for a while. Be patient with your growth, keep up the fantastic work you are doing. The demand for this type of content (building specific trading tools) is slowly but surely growing. I'm sure time will reward your dedication. If you don't mind a suggestion, I would recommend (to accelerate that growth) building tools that are normally only available on paid platforms (Cumulative Volume Delta indicators, footprint charts, market profiles, etc). Many new traders are always looking for free platforms to use them and they will definitely stumble across your videos if they do. Cheers, mate.
Savitzky-Golay filter is not as useful as it is presented here; there's a future leak. Under the hood, Savitzky-Golay calculates a value using a "data window". You chose the value 5 in the savgol_filter method, so your data window length is 5. This means it slides through your data with a window of size 5, fits a polynomial function to ALL 5 points, then outputs the savgol_filter value for the middlemost point. i.e, it uses 2 data points from the future. This is important, obviously, because when using this in real time, you don't know the future... so, you will not catch turns until 2 bars after they happen. This is no issue if you are swinging or otherwise do not need hyper accurate entry. However, there are smoothing methods besides Savitzky-Golay which have online/realtime capabilities.
Your post is very good! In your example, the fact of calculating trend lines based on the last 30 candles limits the temporal extension of the lines drawn. This means that support and resistance lines will be determined primarily based on the most recent trend, not taking into account more distant historical behavior. If we want to identify long-term trends or capture broader historical patterns, wouldn't it be appropriate to consider a more extensive data set when calculating support and resistance lines? How does the algorithm behave?
Hi Chad, thanks so much for your amazing videos, they are so so helpful!! In case if it's interesting for you, would be great to know how to detect sideways market regime. In sideways filtering out signals by trend gives bad effect. But how to detect it early - that's so hard... Which technique might fit for this task? Thank you! All the best.
I got it working but with yfinance data and some tweaks. The only thing not working is the video due to: AttributeError: module 'matplotlib' has no attribute 'subplots' for: fig, ax = plt.subplots()
Is anyone going to talk about the double-look-ahead bias in the video? First Savitzky Golay and then find_peaks... each of those use next data points to calculate previous ones. This is the worst mistake possible to make in technical analysis, and it comes from lack of research. Whether or not the demonstrator in the video is aware to this problem, not stating the lookahead bias is very bad.
He stated at the beginning. This is not a signal in and of itself. It is only intention of showing if in an uptrend or downtrend as TA defines by higher high and higher low, etc.
not that useful from the intro alone you see it repaints a lot. so it changes the past labels past on more info so you have a massive data leakage in your code. for example let's say you go long based on something in combination with the code saying there is an uptrend, now new information comes in and the code changes that specific uptrend label to a downtrend label. now what do you do. should never go back in time and change the labels it gave edit: tldr it uses data from the future to label the past
Hey Oliver! Thanks for watching. I agree with your analysis of the code, it does re-paint the past. I don't think that's a bad thing though, future price action can and should effect our interpretation of past trends
hi, thanks this is really interesting. i am getting following error when displaying in animation. can you pls suggest how to solve it File "C:\Python\trend2.py", line 213, in video = HTML(anim.to_html5_video()) ^^^^^^^^^^^^^^^^^^^^^ File "C:\Python\Lib\site-packages\matplotlib\animation.py", line 1284, in to_html5_video Writer = writers[mpl.rcParams['animation.writer']] ~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Python\Lib\site-packages\matplotlib\animation.py", line 148, in __getitem__ raise RuntimeError(f"Requested MovieWriter ({name}) not available") RuntimeError: Requested MovieWriter (ffmpeg) not availabl
Keep up the work good Chad! though isn't quite practical but it's a nice and refreshing concept from the usual TA. Love to see more stuff in the future.
You mean code it line by line like some of my other videos? I'm experimenting with different formats right now, but I'll take this feedback on board Thanks for watching!
There is a bug in your code, you assign all four ohlc values from just the single "df.open.astype(float)" column
Oops! Thanks for pointing this out.
Thankfully it doesn't effect the later code too much. Just a few parameters need tweaking
Hi
Can you shate the python code? Thanks
Dude, that is amazing! Do you realize this is the perfect filter to count the Elliot Waves? Brilliant
Mans is leveling up my algo trading skillset
I've been looking for a channel like this for a while.
Be patient with your growth, keep up the fantastic work you are doing.
The demand for this type of content (building specific trading tools) is slowly but surely growing.
I'm sure time will reward your dedication.
If you don't mind a suggestion, I would recommend (to accelerate that growth) building tools that are normally only available on paid platforms (Cumulative Volume Delta indicators, footprint charts, market profiles, etc). Many new traders are always looking for free platforms to use them and they will definitely stumble across your videos if they do.
Cheers, mate.
Truly fascinating; I appreciate your sharing of expertise. I eagerly anticipate further enlightening insights from you.
Hi, thank you for sharing videos and your knowledge. Greetings. Have a great year.
Savitzky-Golay filter is not as useful as it is presented here; there's a future leak. Under the hood, Savitzky-Golay calculates a value using a "data window". You chose the value 5 in the savgol_filter method, so your data window length is 5. This means it slides through your data with a window of size 5, fits a polynomial function to ALL 5 points, then outputs the savgol_filter value for the middlemost point. i.e, it uses 2 data points from the future. This is important, obviously, because when using this in real time, you don't know the future... so, you will not catch turns until 2 bars after they happen. This is no issue if you are swinging or otherwise do not need hyper accurate entry.
However, there are smoothing methods besides Savitzky-Golay which have online/realtime capabilities.
Can you share the source code?, Please
Thanks for the update, but how do I download the source code
Hey Chad, Great video. Is there a link to the code? where we can read it more easily?
Good work man...........
Your post is very good! In your example, the fact of calculating trend lines based on the last 30 candles limits the temporal extension of the lines drawn. This means that support and resistance lines will be determined primarily based on the most recent trend, not taking into account more distant historical behavior.
If we want to identify long-term trends or capture broader historical patterns, wouldn't it be appropriate to consider a more extensive data set when calculating support and resistance lines? How does the algorithm behave?
could you share this code ?
can you publish your code?
great job ❤🌹
can you please make a repo for tutorials and share the repo URL so we could have access to tutatials source codes ?
is there a way to do your programs without binance, maybe yahoo finance I don't have a binance account
Hi Chad, thanks so much for your amazing videos, they are so so helpful!!
In case if it's interesting for you, would be great to know how to detect sideways market regime.
In sideways filtering out signals by trend gives bad effect. But how to detect it early - that's so hard... Which technique might fit for this task?
Thank you! All the best.
Have you thought about making it buy with a 1 ATR stop loss and a 2 ATR TP perhaps
i do not understand how does th function animate() see the definitions of variables defined outside of it (df, bars_in_frame, price... etc) ?
This code seems to have some look ahead bias (repainting) , the labelling of the low / high seems to change as new data comes in.
Indeed it does. You should take care in your backtesting to stop it from looking ahead
@@ChadThackray The use cases are limited in such case , perhaps make a revised video where the indicator is not subject to this ?
@@randomdude79404 Maybe you can make one for me! What are the current limitations?
I got it working but with yfinance data and some tweaks. The only thing not working is the video due to:
AttributeError: module 'matplotlib' has no attribute 'subplots' for: fig, ax = plt.subplots()
23:02 - 23:04 It's repainting. Is that correct?
Yeah this is a repainting indicator. Be careful when backtesting it
nice! is that the same to a channel breakout?
Is anyone going to talk about the double-look-ahead bias in the video? First Savitzky Golay and then find_peaks... each of those use next data points to calculate previous ones. This is the worst mistake possible to make in technical analysis, and it comes from lack of research. Whether or not the demonstrator in the video is aware to this problem, not stating the lookahead bias is very bad.
Interesting. Do you think using indicators that look ahead are always bad?
As long as your backtesting system is robust it should be ok, no?
where is the find peak function?
is the code available somewhere? can be used to trade in real time? thanks
repaint and late signals in my opinion
He stated at the beginning. This is not a signal in and of itself. It is only intention of showing if in an uptrend or downtrend as TA defines by higher high and higher low, etc.
not that useful from the intro alone you see it repaints a lot. so it changes the past labels past on more info so you have a massive data leakage in your code.
for example let's say you go long based on something in combination with the code saying there is an uptrend, now new information comes in and the code changes that specific uptrend label to a downtrend label. now what do you do.
should never go back in time and change the labels it gave
edit: tldr it uses data from the future to label the past
Hey Oliver! Thanks for watching.
I agree with your analysis of the code, it does re-paint the past. I don't think that's a bad thing though, future price action can and should effect our interpretation of past trends
hi, thanks this is really interesting.
i am getting following error when displaying in animation. can you pls suggest how to solve it
File "C:\Python\trend2.py", line 213, in
video = HTML(anim.to_html5_video())
^^^^^^^^^^^^^^^^^^^^^
File "C:\Python\Lib\site-packages\matplotlib\animation.py", line 1284, in to_html5_video
Writer = writers[mpl.rcParams['animation.writer']]
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Python\Lib\site-packages\matplotlib\animation.py", line 148, in __getitem__
raise RuntimeError(f"Requested MovieWriter ({name}) not available")
RuntimeError: Requested MovieWriter (ffmpeg) not availabl
Keep up the work good Chad!
though isn't quite practical but it's a nice and refreshing concept from the usual TA.
Love to see more stuff in the future.
Very good and useful tutorial. Thanks Chad 👍
Enjoying your videos and learning a lot from them.
Thank you for sharing!
Very useful!!👍🏼 This video is the highest high
Thanks for a very useful video!
Great work mate!.
How can I do the same in django and display to user on web page ?
You can ask on Stack Overflow or hire a professional if you want help building something like this
I haven't seen the whole thing yet but I'm sure it will be an excellent video.
Thanks for the support!
please code in your video
You mean code it line by line like some of my other videos?
I'm experimenting with different formats right now, but I'll take this feedback on board
Thanks for watching!