How to create two different legends from multiple plot calls

I have to create a visualization where there are multiple line plots (trend lines/moving averages etc.), and multiple scatter charts.

I have successfully created all the charts in a single plot, however I want the legend of scatter chart to be different from that of the line plots.

I want the legend of lineplots to be at the center and that of scatter charts to be at the upper left corner of the plot.

Code for the 4 scatter charts

#Scatter plot
#Using matplotlib scatter plot
scatter1 = plt.scatter(data=df[df.ABC < 10], x='Date', y='ABC', c = 'cyan' , s = 5, label='less')
scatter2 = plt.scatter(data=df[(df.ABC >= 10) & (df.ABC < 40)], x='Date', y='ABC', c = 'dodgerblue' , s = 5, label='normal')
scatter3 = plt.scatter(data=df[(df.ABC >= 40) & (df.ABC < 60)], x='Date', y='ABC', c = 'orangered' , s = 5, label='good')
scatter4 = plt.scatter(data=df[df.ABC > 60], x='Date', y='ABC', c = 'brown' , s = 5, label='more')

Code for the line plots and a calculated metric (all have to be in the same legend box)

#Empty legend line specifying a calulated metric
plt.plot([], [], ' ', label=f'Points above Target Budget PR = {num}/{den} = {ans}%')

#plot using rolling average
#using seaborn.lineplot()
sns.lineplot( x = 'Date',
             y = 'moving avg',
             data = df,
             color='red',
             linewidth=1.5,
             label = 'Moving Average')

#plot a simple time series plot
#using seaborn.lineplot()
sns.lineplot( x = 'Date',
             y = 'Yield',
             color = 'darkgreen',
             linewidth = 1.5,
             data = df,
             label = 'Yield')

I have tried following method to make multiple legends:-

legend1 = plt.legend((scatter1,scatter2,scatter3,scatter4),['less','normal','good','more'], loc = 'upper left')
plt.gca().add_artist(legend1)
plt.legend(loc = 'center')

What above code does is, it creates a legend for scatter chart in top left corner. However the same points are re-written in the legend that is supposed to contain only line plots and the calculated metric. so the legend for scatter chart is represented twice in the chart.

I tried to do other changes in the code too, but the outcome always leads to 1 legend including all, or 2 legends with scatter chart part repeated, or no legend at all.

• It is typically a better option to manage the data prior to any visualizations.
  1. Use pd.cut to bin the data with categories
  2. Add the rolling mean, and median to the dataframe as columns
  3. Convert the new columns to long-form with pandas.DataFrame.melt
  • It's better, because all of the data is in a single place, and it allows for more easily visualizing the data.
• Visually speaking, unless there are two different y-axes (left / right), it creates a cleaner visualization to have a single legend, and not in the middle of the plot, or otherwise covering markers.

import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt

# create sample data
np.random.seed(2023)
df=pd.DataFrame({'ABC': np.random.uniform(low=0, high=100, size=(100,)),
                 'Date': pd.date_range('2019/07/01', periods=100, freq='SM')})

# create bins with labels for the data
bins = [-np.inf, 10, 40, 60, np.inf]
labels = ['less', 'normal', 'good', 'more']
df['cats'] = pd.cut(df.ABC, bins=bins, labels=labels, ordered=True, right=False)

# add the rolling mean and median as a column
df['MA'] = df.ABC.rolling(10).mean()
df['Yield'] = df.ABC.rolling(10).median()

# convert the rolling columns to a long form
df = df.melt(id_vars=['ABC', 'Date', 'cats'], var_name='Labels')

# plot
fig = plt.figure(figsize=(10, 6))
ax = sns.scatterplot(data=df, x='Date', y='ABC', hue='cats', palette=['cyan', 'dodgerblue', 'orangered', 'brown'])
sns.lineplot(data=df, x='Date', y='value', hue='Labels', palette=['red', 'darkgreen'], ax=ax)

sns.move_legend(ax, bbox_to_anchor=(1, 0.5), loc='center left', frameon=False)

df

           ABC       Date    cats Labels      value
0    32.198830 2019-07-15  normal     MA        NaN
1    89.042245 2019-07-31    more     MA        NaN
2    58.805226 2019-08-15    good     MA        NaN
3    12.659609 2019-08-31  normal     MA        NaN
4    14.134122 2019-09-15  normal     MA        NaN
...
195  60.740918 2023-06-30    more  Yield  61.721929
196  47.481600 2023-07-15    good  Yield  56.303730
197  70.959070 2023-07-31    more  Yield  61.721929
198  11.524271 2023-08-15  normal  Yield  56.303730
199  73.279407 2023-08-31    more  Yield  56.303730

The excellent answers previously posted are mostly specific to the OP problem — I'd like to add a generic solution to the problem of using a legend for each Artist type that is present in a plot.

First we need a function to extract from the handles the Artists of the same type

def split(handles_labels, plot_type):
    import matplotlib

    types = dict(plot=matplotlib.lines.Line2D,
                 scatter=matplotlib.collections.PathCollection,)

    try: plot_type = types[plot_type]
    except KeyError: raise ValueError('Invalid plot type.')

    return zip(*((h, l) for h, l in zip(*handles_labels) if type(h) is plot_type))

Then it's quite easy to produce the plot above...

import matplotlib.pyplot as plt

fig, ax = plt.subplots(layout='constrained')

ax.scatter(9, 4, label='A')
ax.plot((5,7),(4,8), label='B')
ax.scatter(3, 7, label='C')
ax.plot((4,8),(6,5), label='D')

handles_labels = ax.get_legend_handles_labels()

l0 = ax.legend(*split(handles_labels, 'scatter'), loc='upper left')
ax.add_artist(l0)
l1 = ax.legend(*split(handles_labels, 'plot'), loc='upper right')

plt.show()

One way to do this (and what I think is the easier way) is to simply completing the plotting and get the legend with all the 6 entries. Then use get_legend_handles_labels() to get the legends and handles, split them up as you need and create them as separate legends. Note the add_artist is so that you add the first one to the axis. The entire code is below... mostly what you have, removed the empty legend line you have and most of the added code is towards the bottom... Also, Title1 & Title2 are optional and you can use/remove it as it suits you.

df=pd.DataFrame({'ABC':np.random.uniform(low=0, high=100, size=(100,)),
                'Date':pd.date_range('2019/07/01', periods=100, freq='SM')})
#Using matplotlib scatter plot
scatter1 = plt.scatter(data=df[df.ABC < 10], x='Date', y='ABC', c = 'cyan' , s = 5, label='less')
scatter2 = plt.scatter(data=df[(df.ABC >= 10) & (df.ABC < 40)], x='Date', y='ABC', c = 'dodgerblue' , s = 5, label='normal')
scatter3 = plt.scatter(data=df[(df.ABC >= 40) & (df.ABC < 60)], x='Date', y='ABC', c = 'orangered' , s = 5, label='good')
scatter4 = plt.scatter(data=df[df.ABC > 60], x='Date', y='ABC', c = 'brown' , s = 5, label='more')

#Empty legend line specifying a calulated metric
#plt.plot([], [], ' ', label=f'Points above Target Budget PR = {num}/{den} = {ans}%')

#plot using rolling average
#using seaborn.lineplot()
sns.lineplot( x = 'Date',
             y = df.ABC.rolling(10).mean(),
             data = df,
             color='red',
             linewidth=1.5,
             label = 'Moving Average')

#plot a simple time series plot
#using seaborn.lineplot()
sns.lineplot( x = 'Date',
             y = df.ABC.rolling(10).median(),
             color = 'darkgreen',
             linewidth = 1.5,
             data = df,
             label = 'Yield')

h,l = plt.gca().get_legend_handles_labels() ##Get the legend handles and lables

l1 = plt.gca().legend(h[:2],l[:2], loc='center', title='Title 1') ##Plot the seborn lines as the first legend
l2 = plt.gca().legend(h[2:],l[2:], loc='upper right', title='Title 2') ##Plot the seborn lines as the first legend

plt.gca().add_artist(l1) # 2nd legend will erases the first, so need to add it
plt.show()