Table of Contents
Multiple axes and subplots
- Add an additional independent axis to a figure
- Add independent subplots to a figure
Plotly is quite liberal with its axes. We can add as many axes as we wish and position them anyway in the figure. Doing so requires the correct parameters to be set however.
For adding an additional x or y axis, or adding subplots, however, there are quicker ways or helper functions. The fact that plotly gives us freedom though helps explain some of the parameters we need to set below.
Two y Axes
Let's start with a simple plot of the number robbery crimes in Victoria (Australia) versus the unemployment rate over a decade.
yrs = ['2004/05',
'2005/06',
'2006/07',
'2007/08',
'2008/09',
'2009/10',
'2010/11',
'2011/12',
'2012/13',
'2013/14']
# number of reported robberies
robs = [1217.0,
1374.0,
1470.0,
1844.0,
1766.0,
1678.0,
1772.0,
1655.0,
1354.0,
1311.0]
# average employment rate as a percentage
emp = [71.0,
71.5,
72.299999999999997,
73.0,
72.0,
71.900000000000006,
72.799999999999997,
72.400000000000006,
72.0,
71.099999999999994]
data = [
go.Scatter(x = yrs, y=robs, name='Robberies'),
go.Scatter(x = yrs, y = emp, name='Empl Rate')
]
fig = go.Figure(data=data, layout=dict())
iplot(fig)
As we can see, the data sets, the number of robberies and the employment rate, require different scales. On a single y axis, employment rate is inaccurately flat.
To add an additional y axis for the employment rate, there are two steps:
- Add the y-axis
- Bind the employment rate trace to the second y axis.
Adding an axis
Basics
An axis is added to the layout dictionary of a figure. Like the default xaxis and yaxis, it takes a dictionary (see Layout - titles, axes and ticks). The graph objects go.XAxis and go.YAxis can be used.
fig.layout.update(yaxis2 = go.YAxis())
As described in Layout - titles, axes and ticks, by default, the attribute or key for axes in the layout dictionary of a plot are: xaxis and yaxis. That is, to modify the format of an axis, one needs to modify fig.layout.xaxis or fig.layout.yaxis.
Whenever additional axes are to be added, plotly requires that the same attributes be used (ie, yaxis and xaxis) but with the number 2, or 3 etc, for the second or third axis of that kind. For this reason, our second yaxis has to be given to the key yaxis2 in the layout dictionary.
Positioning the new axis
The above code is incomplete and won't work. We need to tell the new axis two things:
- Whether it is free floating or to overlap with the pre-existing axis;
- Which side of the current plot is it to be (top, bottom, left or right).
In our case, we want the new axis to be overlapping. Like mentioned above, plotly lets axes do what ever you want, essentially, so the default is that an axis is free. We have to tell it to be sort of 'connect' with the pre-existing yaxis. The attribute used for this setting is overlaying. Its default is free. To connect it to a pre-existing axis, pass the kind of axis it needs to connect to. If it needs to connect to a y axis, overlaying='y'; if it needs to connect to an x axis, overlaying='x'.
The default side for a y axis, is the left, whilst for an x axis, it is the bottom. Thus, for our new y axis, by default it will literally be placed on top of the pre-existing one, which doesn't work well. For it to be placed on the other side, we must use the attribute side and set it to the opposite of the pre-existing one, which for us, by default is on the left side.
If you'd like, try adding a new axis without these attributes set appropriately to get a an intuitive feeling for what they do.
So, to appropriately position our axis, we must add it with the following attributes:
fig.layout.update(yaxis2 = go.YAxis(overlaying='y', side='right'))
You'll notice though, that neither of the traces have been bound to the new axis. Try zooming or panning the second y axis - the new axis remains unchanged and no trace follows it ...
Binding a trace to a new axis
As mentioned above, plotly is liberal with how the axes are arranged and how traces are bound to them. Traces can be bound to any trace, technically. By default, each trace is bound to the default or initial x and y axes.
To change this, each trace has two attributes, yaxis and xaxis that determine which axis or axes that trace is bound to.
To repeat, these are attributes of each trace and are different from the attributes of the layout dictionary. In the definition of a trace, xaxis defines which x axis the trace will be bound to. In the definition of the layout xaxis defines the formatting and styling of the x axis for that figure.
In our case, we want our second trace to be bound to the second y axis.
Annoying Convention!!
Now this is oddly inconsistent, but even though the second y axis must be called yaxis2, the name we must give in the trace is not this but rather y2. It is shorter, but annoyingly inconsistent. The same, of course, applies to any new x axes.
So, our code for the second trace needs to be ...
go.Scatter(x = yrs, y = emp, name='Empl Rate',
# bind this trace to the second y axis,
# or the y axis defined by the attrbite "yaxis2"
yaxis='y2')
Final
The final code for a second y axis would look something like
data = [
go.Scatter(x = yrs, y=robs, name='Robberies'),
go.Scatter(x = yrs, y = emp, name='Empl Rate',
yaxis='y2') # binding to the second y axis
]
# settings for the new y axis
y2 = go.YAxis(overlaying='y', side='right')
# adding the second y axis
layout = go.Layout(yaxis2=y2)
fig = go.Figure(data=data, layout=go.Layout(yaxis2=y2))
iplot(fig)
Adding some color to the axes to make it clearer which is for which trace is often helpful.
data = [
go.Scatter(x = yrs, y=robs, name='Robberies'),
go.Scatter(x = yrs, y = emp, name='Empl Rate', yaxis='y2')
]
# Add titles and color the font of the titles to match that of the traces
# 'SteelBlue' and 'DarkOrange' are the defaults of the first two colors.
y1 = go.YAxis(title='Robberies per yr', titlefont=go.Font(color='SteelBlue'))
y2 = go.YAxis(title= 'Employment Rate', titlefont=go.Font(color='DarkOrange'))
# update second y axis to be position appropriately
y2.update(overlaying='y', side='right')
# Add the pre-defined formatting for both y axes
layout = go.Layout(yaxis1 = y1, yaxis2 = y2)
fig = go.Figure(data=data, layout=layout)
iplot(fig)
Two subplots
If the two traces were in their separate subplots, then we wouldn't need to worry about accommodating two scales.
There is a helper function that does the heavy lifting for generating subplots for us. (It's useful to remember though that you could do it all yourself just by manipulating the positioning and binding of various axes.)
make_subplots() function
Import
We will need to import the helper function from plotly.tools.
from plotly.tools import make_subplots
Initialising the figure for subplots
make_subplots prepares a figure dictionary that is set up to have subplots.
So, unlike for previous plots, if you want subplots, you will tend to start by defining your figure graph object or dictionary and then define your traces etc.
make_subplots expects the subplots to be arranged in a grid or table, with a certain amount of rows and columns, which you have to specify.
Thus, it's main options or key-word arguments are:
rows- number of rows of subplotscols- number of columns of subplots
For our case, we will only have two subplots. Let's arrange them in a single row, with two columns for each of the subplots.
fig = make_subplots(rows=1, cols=2)
The guide output
Note - make_subplots returns a handy summary of the grid of subplots we've just initialised ...
This is the format of your plot grid:
[ (1,1) x1,y1 ] [ (1,2) x2,y2 ]
- Each square brack -
[]- represents a single subplot - the numbers in round brackets -
(1, 2)- represents which row and column the subplot belongs to. Ie(row, col). - The
x1, y1tells us which of the axes are bound to each subplot. Recall from above thatyaxis2is referred to asy2,yaxis3asy3, etc. These references are what we see here. You'll notice that there are new axes for each subplot. Ie, the second subplot hasx2, y2You don't really need to change these unless you're doing some serious customising. But it is helpful to see that all this function is doing is the same sort of stuff we were doing above.
DIGGING DEEPER - The figure object
It's worthwhile, if you're interested, at this point to see what our figure dictionary looks like, keeping in mind that all make_subplots() has done is write out this particular figure dictionary for us.
pply(fig)
(see Methods for updating the figure ... for using pply())
returns ...
{ 'data': [],
'layout': { 'xaxis1': { 'anchor': 'y1', 'domain': [0.0, 0.45]},
'xaxis2': { 'anchor': 'y2', 'domain': [0.55, 1.0]},
'yaxis1': { 'anchor': 'x1', 'domain': [0.0, 1.0]},
'yaxis2': { 'anchor': 'x2', 'domain': [0.0, 1.0]}}}
What has been done:
- separate x and y axes were added to the layout for each subplot
- The two pairs of axes were bound to each other using
'anchor': 'y1'. - Each axis was given a portion of the whole figure using
'domain': [0.0, 0.45]. This means that the axis spans, in this example, from the bottom left to 45% of the way to the right. FOr y axes,0.0corresponds to the top left.
Note - after using make_subplots(), we can change these default settings to customise the spatial arrangement of our plot.
Append traces
To add a particular trace to a particular subplot, use fig.append_trace(trace, row, col).
This is a function built into figure graph objects. You pass into the function the trace you want append, and then the row and column coordinates of the subplot you want to add the trace to. This is what the guide output of the make_subplots() function is really useful for.
For us, we'll append the robberies trace to the first subplot (1, 1), and the employment rate trace to the second subplot (1, 2).
fig.append_trace(go.Scatter(x = yrs, y=robs, name='Robberies'),
1, 1) # first row, first column
fig.append_trace(go.Scatter(x = yrs, y = emp, name='Empl Rate'),
1, 2) # first row, second column
iplot(fig)
With the subplot helper functions, this is easier than adding a new axis.
It's worthwhile, again, if you're interested, looking at the figure dictionary again, to see how the traces were 'appended' to their subplots
pply(fig)
returns ...
{ 'data': [ { 'name': 'Robberies',
'type': 'scatter',
'xaxis': 'x1', # This how the trace is bound
'yaxis': 'y1'}, # This how the trace is bound
{ 'name': 'Empl Rate',
'type': 'scatter',
'xaxis': 'x2', # This how the trace is bound
'yaxis': 'y2'}], # This how the trace is bound
'layout': { 'xaxis1': { 'anchor': 'y1', 'domain': [0.0, 0.45]},
'xaxis2': { 'anchor': 'y2', 'domain': [0.55, 1.0]},
'yaxis1': { 'anchor': 'x1', 'domain': [0.0, 1.0]},
'yaxis2': { 'anchor': 'x2', 'domain': [0.0, 1.0]}}}
You'll notice that all 'appending' involved was binding the trace to the appropriate axes in the same way we were binding a trace to a new y axis above. Also notice that the appropriate axes for each subplot are those that were listed in the guide ouput of the make_subplots() function.
Share Axes
With our two data sets, one axis is identical for them - time. This is how we did the first plot with two y axes ... both y axes shared the same x axis.
We do this by using the shared_xaxes=True and/or shared_yaxes=True option or key-word argument of the make_subplots() function.
fig = make_suplots(rows=r, cols=c, shared_xaxes=True)
For our two subplots, and any subplots for that matter, x or y axes can be shared also.
So, for us, we'll rearrange our subplots into a single column, so that they can share an x axis of time.
Initialise the figure
# two subplots, each in a row, forming one column
fig = make_subplots(rows=2, cols=1,
shared_xaxes=True) # sharing the x axis.
returns ...
This is the format of your plot grid:
[ (1,1) x1,y1 ]
[ (2,1) x1,y2 ]
Note - there are two y axes (y1 and y2) but only one x axis, that is bound to both suplots.
Append traces
Just as before, but with the rows and cols swapped.
fig.append_trace(go.Scatter(x = yrs, y=robs, name='Robberies'), 1,1)
fig.append_trace(go.Scatter(x = yrs, y = emp, name='Empl Rate'), 2, 1)
Figure dictionary
It's worthwhile looking at the figure dictionary again.
{ 'data': [ { 'name': 'Robberies',
'type': 'scatter',
'xaxis': 'x1',
'yaxis': 'y1'},
{ 'name': 'Empl Rate',
'type': 'scatter',
'xaxis': 'x1',
'yaxis': 'y2'}],
'layout': { 'xaxis1': { 'anchor': 'y2', 'domain': [0.0, 1.0]},
'yaxis1': { 'anchor': 'free',
'domain': [0.575, 1.0],
'position': 0.0},
'yaxis2': { 'anchor': 'x1', 'domain': [0.0, 0.425]}}}
What was done differently to share an axis:
- Both traces were bound to the same x axis
- No second x axis was defined
- The first y axis, which is for the top subplot, was not bound to any x axis, but left free (
'anchor': 'free').
Exercise
In the data below, you have both the life expectancy and GDP per capita for Germany from 1900 to the present.
- Plot both of these data sets on one plot with two axes, one for each data set.
- Label and style the axes in a way that makes them distinguishable.
- Plot both of the data sets on two subplots.
- As both data sets share a dimension - time - share an axis between the two subplots.
years = [1900, 1901, 1902, 1903, 1904, 1905, 1906, 1907, 1908, 1909, 1910,
1911, 1912, 1913, 1914, 1915, 1916, 1917, 1918, 1919, 1920, 1921,
1922, 1923, 1924, 1925, 1926, 1927, 1928, 1929, 1930, 1931, 1932,
1933, 1934, 1935, 1936, 1937, 1938, 1939, 1940, 1941, 1942, 1943,
1944, 1945, 1946, 1947, 1948, 1949, 1950, 1951, 1952, 1953, 1954,
1955, 1956, 1957, 1958, 1959, 1960, 1961, 1962, 1963, 1964, 1965,
1966, 1967, 1968, 1969, 1970, 1971, 1972, 1973, 1974, 1975, 1976,
1977, 1978, 1979, 1980, 1981, 1982, 1983, 1984, 1985, 1986, 1987,
1988, 1989, 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009,
2010, 2011, 2012, 2013, 2014, 2015]
le_ge = [ 43.915 , 44.222 , 44.529 , 44.836 ,
45.143 , 45.45 , 46.04166667, 46.63333333,
47.225 , 47.81666667, 48.40833333, 49. ,
49.59166667, 50.18333333, 46.17648422, 40.52556325,
38.9888193 , 40.08037644, 32.93682031, 48.32065914,
53.5 , 55.01820257, 55.62354801, 56.22889344,
56.83423887, 57.4395843 , 57.85150116, 58.26341802,
58.67533488, 59.08725174, 59.4991686 , 59.91108546,
60.32300232, 60.73491918, 61.14683604, 61.5587529 ,
61.84933643, 62.13991996, 62.43050348, 61.07442034,
60.73821096, 59.08225406, 55.08617092, 49.79008778,
37.09400464, 29.0979215 , 60.61319836, 62.21527522,
63.81735208, 65.41942894, 67.0215058 , 67.18742266,
67.51033952, 67.82125638, 68.12117324, 68.4080901 ,
68.70345102, 68.62532856, 69.36929231, 69.48021979,
69.40190727, 69.99702797, 70.16889372, 70.26131586,
70.82344196, 70.81075623, 70.92828395, 71.15404398,
70.80345367, 70.65682236, 70.9 , 71. ,
71.2 , 71.5 , 71.8 , 71.9 ,
72.3 , 72.6 , 72.7 , 72.9 ,
73.1 , 73.4 , 73.6 , 74. ,
74.4 , 74.6 , 74.8 , 75.1 ,
75.3 , 75.4 , 75.4 , 75.6 ,
75.9 , 76.2 , 76.4 , 76.6 ,
76.9 , 77.3 , 77.7 , 77.9 ,
78.1 , 78.3 , 78.5 , 78.8 ,
79.1 , 79.4 , 79.7 , 79.8 ,
80. , 80. , 80.2 , 80.3 ,
80.5 , 80.7 , 80.9 , 81.1 ]
gdp_ge = [ 4596, 4422, 4457, 4634, 4751, 4783, 4858, 5002, 5017,
5049, 5162, 5277, 5477, 5692, 4793, 4559, 4633, 4679,
4746, 4131, 4482, 4969, 5415, 4500, 5270, 5857, 6015,
6618, 6909, 6884, 6791, 6276, 5809, 6178, 6738, 7232,
7852, 8305, 8894, 9674, 9711, 10310, 10407, 10722, 11120,
8284, 4084, 4504, 5258, 6112, 7251, 7884, 8561, 9252,
9926, 10998, 11751, 12385, 12884, 13759, 14808, 15317, 15872,
16221, 17100, 17838, 18262, 18311, 19254, 20409, 21218, 21695,
22497, 23461, 23662, 23630, 24904, 25678, 26444, 27515, 27765,
27846, 27645, 28227, 29135, 29851, 30514, 30986, 31906, 32706,
31476, 32844, 33221, 32689, 33375, 33843, 34008, 34578, 35254,
35931, 36953, 37517, 37458, 37167, 37614, 37901, 39352, 40693,
41199, 38975, 40632, 42080, 42959, 42887, 43444, 44053]
Exercise 2 (extension)
Below is the same data as for exercise 1, but for France.
- In your two subplots from above, add the french data.
- Use the following trace attributes to bind the two countries accross your subplots:
- The colors of the lines of each trace can be made identical for the same country (hint:
line = go.Line()). legendgroup- takes a string, and each trace with an identical string will be grouped together in the legend. On clicking, the whole group is toggled together too.showlegend- boolean (True/False). For each trace, determines whether there is a legend item appearing.
- The colors of the lines of each trace can be made identical for the same country (hint:
le_fr = [ 45.08 , 47.01 , 48.01 , 48.43 , 48.08 , 48.36 ,
47.74 , 48.28 , 49.3 , 50.01 , 51.37 , 48.17 ,
51.62 , 51.35 , 37.85 , 35.63 , 39.51 , 42.6 ,
34.34 , 47.52 , 51.6 , 52.6968, 54.9336, 54.6704,
55.2972, 54.424 , 54.0708, 55.8476, 55.4944, 54.3312,
56.938 , 57.0048, 57.3416, 57.7884, 58.4552, 58.422 ,
58.9188, 59.2856, 59.0924, 59.7492, 49.586 , 57.8128,
57.5896, 53.4864, 47.3532, 55.13 , 62.5568, 64.1636,
66.0204, 65.1172, 66.594 , 66.3308, 67.6276, 67.5644,
68.4412, 68.708 , 68.7448, 69.1816, 70.4184, 70.4552,
70.672 , 71.2588, 70.7956, 70.6524, 71.6192, 71.456 ,
71.8728, 71.8696, 71.8664, 71.6032, 72.5 , 72.6 ,
72.8 , 73.1 , 73.3 , 73.2 , 73.4 , 73.8 ,
74.1 , 74.3 , 74.5 , 74.8 , 75. , 75.2 ,
75.5 , 75.7 , 76. , 76.4 , 76.6 , 76.9 ,
77.1 , 77.3 , 77.5 , 77.7 , 77.9 , 78.1 ,
78.4 , 78.7 , 78.8 , 78.9 , 79.1 , 79.2 ,
79.4 , 79.7 , 80.1 , 80.4 , 80.7 , 80.9 ,
81. , 81. , 81.2 , 81.4 , 81.6 , 81.7 ,
81.8 , 81.9 ]
gdp_fr = [ 4314, 4100, 4048, 4181, 4261, 4367, 4304, 4648, 4629,
4774, 4542, 5004, 5420, 5373, 4506, 3968, 4548, 4506,
3861, 4258, 4550, 4326, 5188, 5434, 6032, 6075, 6284,
6157, 6522, 7078, 6835, 6534, 5962, 6140, 5929, 5785,
5869, 6008, 5861, 6106, 4821, 4666, 4700, 4691, 3594,
4423, 5909, 6228, 6988, 7367, 7914, 8301, 8446, 8622,
9006, 9453, 9907, 10442, 10681, 10911, 11642, 12168, 12767,
13235, 13969, 14514, 15158, 15759, 16321, 17339, 18185, 18891,
19570, 20486, 20997, 20851, 21661, 22270, 22928, 23647, 23962,
24186, 24753, 25188, 25497, 25917, 26453, 26963, 28101, 28942,
29476, 29707, 30033, 29719, 30303, 30823, 31141, 31756, 32764,
33707, 34774, 35197, 35333, 35371, 36090, 36395, 37001, 37641,
37505, 36215, 36745, 37328, 37227, 37309, 37218, 37599]