mpf and subplots

[rman58]

Hi Daniel!
Thank you for your work to improve mplfinance!
How can I use MPF in old code (I write it by analogy):
fig, (ax1, ax2, ax3) = plt.subplots(nrows=3, ncols=1)
mpf.plot(df, type='candle', ax=ax1)

[DanielGoldfarb]

This is the second request I've gotten for something like this, so will probably do something about it. There are some issues though. I'll describe here what I am thinking so that people can comment.

mplfinance.plot() does a lot for you automatically. One of the aspects of the code, that allows mplfinance to do so, is that mplfinance.plot() owns the Figure and all Axes. Passing in an Axes object would limit what plot() can do, since it would then not own the Figure and Axes.

Also, passing in an Axes object would effectively make plot() a lower level function, so that in order for plot() to maintain its current capabilities, along with a more limited set of capabilities for the ax=Axes case, the code would become needlessly complex, branching this way and that depending on whether the caller passed in an Axes instance or not.

I am inclined instead to create a new API, similar to plot() but somewhat limited in its capabilities. I imagine something like this:

mpf.axplot(ax, df, type, upcolor, downcolor)

axplot() would do the following:

• plot either 'ohlc' or 'candle' on the given Axes
• automatically format the date axis (hours/minutes versus months/years, etc).
• allow customization of up/down candle and ohlc colors

None of the other features of plot() would be available to axplot() -- no volume, no customization of appearance (other than up/down colors of candles and ohlc), no saving the plot as a file, etc. All of those things can be done by the caller, since the caller has already decided to write lower level code by instantiating and using instances of Figure and Axes.

I am also thinking maybe to restructure the code in such a way that plot() would itself call axplot(). If I can do that, it should make the code easier to maintain -- instead of having to maintain similar code in potentially two different places: both in plot() and in axplot().

[schorschie]

After your explanation I think it would be better to leave the plot() method as is, to keep it's capabilities.
Maybe instead of creating a new api, just return the figure handle of the plot?

[DanielGoldfarb]

I will consider providing a keyword option to return the Figure. There are potentially problems with this: by maintaining ownership of the Figure and Axes, mpf.plot() is able to do a lot of stuff automatically for the caller. That is part of the philosophy behind mpf.plot() being a "higher level" API compared with the old mpl-finance or the proposed axplot().

So if anything, for now, I will consider an option, maybe noshow=True which means: (1) don't use context manager, (2) don't call plt.show(), and (3) return Figure. It is also almost certain at this point that mpf.axplot() will be written; while it may not be the ideal solution for you, there are definitely others who have expressed an interest in such an API.

Please feel free, everyone, to continue this discussion. My ideal hope for mplfinance is that it can make it easy for people to code 75% to 85% of the use-cases they have (and only slightly more complicated to code the rest). Contributing to these discussions is a great way to help us achieve that goal.

Also, please consider, just as a test/proof of concept, cloning the code and modifying it to return the Figure and then, most importantly, please let me know if that works for you, or if perhaps you come up against another issue that we need to solve. If I know that it easily does for you want you want, that will weigh strongly in favor of me adding the feature to return the Figure object. Thank you.

[verata-veritatis]

Hi Daniel;

I think the ability to add a subplot or a custom indicator to the plot is extremely important, especially for algorithm developers. I create unique indicators all the time and not having a subplot ability would be quite detrimental. In fact, I found this thread after attempting to plot a variation of RSI with mplfinance without success.

[DanielGoldfarb]

@verata-veritatis ... I just want to clarify a few things on my end, and then ask for some clarification as to what you want to see.

At present, via mpf.plot(), there are two subplots available: the "main panel" that shows the OHLC bars/candlesticks, and the "lower panel" which can be used to display volume and/or other data. Please make sure you have read through the "additional plot" documentation to see how you can display your RSI indicator on the "lower panel".

As it is now, if you display both volume and an indicator on the lower panel, mpf.plot() will automatically display the indicator on a secondary y-axis.

This feature will be enhanced for the next release of mplfinance as follows: a secondary y-axis will be available for both the main and lower panel, and you will be able to specify for each addplot() call whether you want to plot to be on the primary or secondary y-axis.

Now the question is, after reading the "additional plot" documentation are you able to display your RSI indicator? Is there something else you need (if so, what)? Do you have a specific need for more subplots(), and if so for what?

Here is my current thinking on additional subplots: If we implement mpf.axplot() as described above, then you will be able to make as many subplots as you want, and have the freedom to organize them however you want. However it means you have to do a lot of the matplotlib work yourself to structure those plots. This situation won't be much different that the old api (for example, calling candlestick_ohlc()).

We may also make one more subplot avabilable via mpf.plot() (one more than the two that already exist, so a total of three subplots available) and perhaps allow two possible layouts for these subplots. Here are some pictures to clarify:

---

These two layouts are available now:

• single main panel
  

• two panels: main and lower
  

---

In addition we would make the following layouts available:

• three panels: main with two below
  

---

• three panels: main with one above and one below
  
  **

---

---

Please let me know if the two panel case (the one that exist now: see "additional plot" documentation) is adequate for what you want to do. If not, please explain clearly why it's not adequate.

Also, if the existing two panel case will not work, indicate whether or not you think the above proposal (of a few different layouts) or something similar would work. Thank you!

[zillionare]

Hi Daniel,

Thanks for the effort. I personally would like to see if mpf have the ability to draw subplots. The reason: I need to draw candlestick for at least two timeframe( day and 30min), and a corresponding index candlestick.

[DanielGoldfarb]

@hbaaron please propose how you would want the mpf interface to look/work to do that. Thanks.

[DanielGoldfarb]

double y-axis is now released. See the "additional plot" documentation here.

I am currently experimenting with a couple different ways of implementing subplots to see which is going to work the best.

[pkgvrpdm]

It will be greatly appreciated if many subplots are available in this lib.
One more question can we handle realtime tick by tick data updates in it?

[DanielGoldfarb]

@pkgvrpdm

... can we handle realtime tick by tick data updates in it?

please check this issue and confirm whether that covers what you are looking for.