Overlayed candle plots - set color with make_addplot

[PT-BalazsBorbola]

We're trying to make a chart with multiple, overlapping candlesticks. The code we're using is this:

    for idx, dataset in enumerate(ohlc_collector):
        ohlc_data = dataset[1]
        apds.append(mpf.make_addplot(ohlc_data, type="candle", secondary_y=False))
        fig, axlist = mpf.plot(ohlc_data, addplot=apds, type="candle", show_nontrading=True, 
                               figratio=(10, 5), ylabel="", savefig="Chart_OHLC.pdf", returnfig=True)

Using this code, the end result looks like this:

Quite ugly, I know. It will get tidied up and with better resolution it will work out well enough. What we need however, is to color every candlestick data series a SINGLE color (unlike other candlesticks; basically one series red, another one blue, and so on), not unlike a multiline chart.

However, you can't use color= for make_addplot() if type is "candle", and if you apply a style at plot(), it will apply to ALL candlesticks. How could you color each and every candlestick a different color?

[DanielGoldfarb]

Candle colors are controlled by mplfinance styles, specifically through the marketcolors attribute of an mplfinance style.

There are two approaches to accomplishing what you want. The best is to modify mplfinance.make_addplot() to accept a marketcolors kwarg. This will allow to use different candle colors for each addplot.

The change is fairly easy; about 10 lines of code in src/mplfinance/plotting.py. I've made the changes on my side and they appear to work. However I am also in the midst of a bunch of other stuff so I won't be able to push the changes out for several weeks or more. Therefore I will post the changes here and you can make these changes to your own version of mplfinance:

 git diff -w

diff --git a/src/mplfinance/plotting.py b/src/mplfinance/plotting.py
index 21f6945..9eecd52 100644
--- a/src/mplfinance/plotting.py
+++ b/src/mplfinance/plotting.py
@@ -858,7 +858,14 @@ def _addplot_collections(panid,panels,apdict,xdates,config):
     if not isinstance(apdata,pd.DataFrame):
         raise TypeError('addplot type "'+aptype+'" MUST be accompanied by addplot data of type `pd.DataFrame`')
     d,o,h,l,c,v = _check_and_prepare_data(apdata,config)
-    collections = _construct_mpf_collections(aptype,d,xdates,o,h,l,c,v,config,config['style'])
+    mc = apdict['marketcolors']
+    if isinstance(mc,dict):
+        apstyle = config['style'].copy()
+        apstyle['marketcolors'] = mc
+    else:
+        apstyle = config['style']
+
+    collections = _construct_mpf_collections(aptype,d,xdates,o,h,l,c,v,config,apstyle)

     if not external_axes_mode:
         lo = math.log(max(math.fabs(np.nanmin(l)),1e-7),10) - 0.5
@@ -1088,6 +1095,10 @@ def _valid_addplot_kwargs():

         'stepwhere'    : { 'Default'     : 'pre',
                            'Validator'   : lambda value : value in valid_stepwheres },
+
+        'marketcolors' : { 'Default'     : None, # use 'style' for default, instead.
+                           'Validator'   : lambda value: isinstance(value,dict) },
+
     }

     _validate_vkwargs_dict(vkwargs)

---

After you make the above changes, you can then call mplfinance as in the following example:

aapldf  = pd.read_csv('data/yahoofinance-AAPL-20040819-20180120.csv',index_col=0,parse_dates=True).iloc[-61:-1]
googdf = pd.read_csv('data/yahoofinance-GOOG-20040819-20180120.csv',index_col=0,parse_dates=True).iloc[-61:-1]

mcblue  = mpf.make_marketcolors(base_mpf_style='default',up='b',down='b',ohlc='b')
mcgreen = mpf.make_marketcolors(base_mpf_style='default',up='limegreen',down='limegreen',ohlc='limegreen')

sblue = mpf.make_mpf_style(base_mpf_style='default',marketcolors=mcblue)

ap = mpf.make_addplot(googdf,type='candle',marketcolors=mcgreen)
mpf.plot(aapldf,type='candle',style=sblue,returnfig=True,addplot=ap)

The result looks like this:

Notice that ...

Notice that when the prices of the two time-series are different orders of magnitude (as in the above example) then two y-axis scales are used: one on the left and one on the right. Be careful if you have more than two time-series of various orders of magnitue, since a matplotlib plot can only easily have two y-axes of different orders of magnitude. (There are ways to have more than two y-axes, but it is tricky to label them unambiguously).

Notice also that when the candles for a given time-series are all one color, then you cannot tell which candles are up (close higher than open) and which are down. Alternatively you can use `type='ohlc':

---

If you don't want to make the above change to your own installed version of mplfinance (until we can release the change officially), there is a kludgey workaround you can implement by calling mplfinance twice: once to have it return the Axes objects, and a second time passing the same Axes objects back into mplfinance. I don't recommend doing this, but it does allow you to accomplish the same thing without modifying mplfinance:

aapldf  = pd.read_csv('data/yahoofinance-AAPL-20040819-20180120.csv',index_col=0,parse_dates=True).iloc[-61:-1]
googdf = pd.read_csv('data/yahoofinance-GOOG-20040819-20180120.csv',index_col=0,parse_dates=True).iloc[-61:-1]
datalen = len(googdf)
if datalen != len(aapldf):
    raise ValueError('len(googdf) != len(aapldf)')

mcblue  = mpf.make_marketcolors(base_mpf_style='default',up='b',down='b',ohlc='b')
mcgreen = mpf.make_marketcolors(base_mpf_style='default',up='limegreen',down='limegreen',ohlc='limegreen')

sblue = mpf.make_mpf_style(base_mpf_style='default',marketcolors=mcblue)
sgreen = mpf.make_mpf_style(base_mpf_style='default',marketcolors=mcgreen)

# Create a dummy ohlc dataframe with only a single data point and the rest `nan` values
# (mplfinance will reject a dataframe that is 100% nan values, thus the need for one data point).
# The dummy ohlc df will be used in the
# - first `mpf.plot()` call, for the addplot, in order to create a secondary y axis (`axlist[1]`) below), and in the
# - second `mpf.plot()` as an "empty" df, so as not to overwrite what was plotted in the first `mpf.plot()` call.
df = pd.DataFrame()
df['Open']  = [float('nan')]*datalen
df['High']  = [float('nan')]*datalen
df['Low']   = [float('nan')]*datalen
df['Close'] = [float('nan')]*datalen
df.index = aapldf.index.copy()
# Add one data point to the empty data frame 
# using the data that will be plotted by the `addplot` kwarg (i.e. on the secondary axis)
# (notice we are setting O,H,L, and C all to the 'Open' value so that this candle or bar will be barely visible)
df.loc[df.index[0]] = googdf.loc[df.index[0],'Open']

# Now plot the first candle set, creating a secondary y axis (with the dummy df) for the second candle set:
# set `returnfig=True` to get back the Axes list to be used in the second `mpf.plot()` call:
ap = mpf.make_addplot(df,type='candle',secondary_y=True)
fig, axlist = mpf.plot(aapldf,type='candle',style=sblue,returnfig=True,addplot=ap)

# Now change the single point to the value that was used for the first set of candles 
# so that it matches the order of magnitude of the primary axis:
df.loc[df.index[0]] = aapldf.loc[df.index[0],'Open']

# Now plot the secondary axis data, with it's own style
# while the primary axis gets the dummy dataframe:
ap = mpf.make_addplot(googdf,type='candle',ax=axlist[1],secondary_y=True)
mpf.plot(df,type='candle',ax=axlist[0],style=sgreen,addplot=ap)

# Now display or save the plot:
if display_the_plot:
    mpf.show()
elif save_the_plot:
    fig.savefig('plotname.png')

[PT-BalazsBorbola]

Thank you for the detailed response! I went with option #1 and modified mplfinance. It was easy enough and it worked like a charm. I really hope this change is going to make it into an upcoming release. Thanks again for the answer!

[DanielGoldfarb]

I went with option #1 and modified mplfinance. It was easy enough and it worked like a charm.

Glad to hear it. Thank you for letting me know. It will definitely be released in the next release or so, just there probably won't be another release for at least four to six weeks (due to other work taking priority).

All the best. --Daniel

[DanielGoldfarb]

Another thing I was thinking ...

Since in the above example of market colors, the color is used to distinguish one series from another, and so it is not possible to distinguish up candles from down colors ...

I don't know if the following fits for your application, but it is possible to set your marketcolors in such a way that the center of each candle shows one of two colors for up or down (perhaps black and white) while the candle edges and wicks can be set to a color that distinguishes the series (blue, green, red, etc).

(You can also, when doing this, make the edges and wicks thicker if that helps. See this tutorial about "widths" and in particular search for "candle_linewidth" within the tutorial).

[PT-BalazsBorbola]

Yes, that's basically the next thing to do, making the chart more informative and tidying it up a little. Luckily enough, our dataset contains at most 5 or 6 entries a day, and typically about 1 to 3. This means that ups and downs are usually quite straightforward, but in some cases this extra will be useful. We've already addes a legend manually for the different colors, as that is necessary to distinguish/identify each of the series. What could be useful for such a scenario though, is something like a helper function that creates the matplotlib legend colors based on the marketcolors. Doing this for a single color is straightforward enough, but for stuff like distinct edges and wicks it gets complicated pretty fast. This is just an idea though; right now for our specific case I believe we already have more than what we askes for :)

…

On 2021. november 15. 17:36:19 Daniel Goldfarb ***@***.***> wrote: Another thing I was thinking ... Since in the above example of market colors, the color is used to distinguish one series from another, and so it is not possible to distinguish up candles from down colors ... I don't know if the following fits for your application, but it is possible to set your marketcolors in such a way that the center of each candle shows one of two colors for up or down (perhaps black and white) while the candle edges and wicks can be set to a color that distinguishes the series (blue, green, red, etc). (You can also, when doing this, make the edges and wicks thicker if that helps. See this tutorial about "widths" and in particular search for "candle_linewidth" within the tutorial). — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[DanielGoldfarb]

#471 ( Thank you @miya779 ) Released to production on Dec 15, 2021 ( mplfinance version 0.12.8b6 )