We'd like to try making ternary plots in Python. We'll use python-ternary. I'm following the examples from the GitHub repo.
The ternary package can be installed with (you will need to do this in a cell in Google Colab, for example):
pip install python-ternary
After doing that, check it worked:
In [1]:
import ternary
In [2]:
uid = "1r7AYOFEw9RgU0QaagxkHuECvfoegQWp9spQtMV8XJGI"
url = f"https://docs.google.com/spreadsheets/d/{uid}/export?format=csv"
In [3]:
import pandas as pd
df = pd.read_csv(url) # Line 1.
In [4]:
df.head()
Out[4]:
In [5]:
df.describe()
Out[5]:
In [6]:
df.groupby('Facies Association').count()
Out[6]:
...and we can make cross-plots of the numeric parameters:
In [7]:
import seaborn as sns
sns.pairplot(df, vars=['Matrix', 'Quartz', 'Carbonate', 'Bioclasts', 'Authigenic'], hue='Facies Association')
Out[7]:
The data need to be normalized for a ternary plot. We'll normalize to 100:
In [8]:
cols = ['Carbonate', 'Quartz', 'Matrix']
for col in cols:
df[col[0]] = df[col] * 100 / df[cols].sum(axis=1)
In [9]:
df.head()
Out[9]:
In [10]:
fig, tax = ternary.figure(scale=100)
fig.set_size_inches(5, 4.5)
tax.scatter(df[['M', 'Q', 'C']].values)
tax.gridlines(multiple=20)
tax.get_axes().axis('off')
Out[10]:
In [11]:
# Set up the figure.
fig, tax = ternary.figure(scale=100)
fig.set_size_inches(10, 9)
# Plot points.
tax.scatter(df[['M', 'Q', 'C']].values)
# Axis labels. (See below for corner labels.)
fontsize = 14
offset = 0.08
tax.left_axis_label("Carbonates %", fontsize=fontsize, offset=offset)
tax.right_axis_label("Quartz %", fontsize=fontsize, offset=offset)
tax.bottom_axis_label("Matrix %", fontsize=fontsize, offset=-offset)
tax.set_title("Sobrarbe Formation", fontsize=20)
# Decoration.
tax.boundary(linewidth=1)
tax.gridlines(multiple=10, color="gray")
tax.ticks(axis='lbr', linewidth=1, multiple=20)
tax.get_axes().axis('off')
tax.show()
In [12]:
df.head()
Out[12]:
Let's colour the points by the Facies Association.
And let's compute the mean for each of the columns of interest too.
In [13]:
mean= df[['M', 'Q', 'C']].mean(axis=0)
mean
Out[13]:
Now we can make a plot:
In [14]:
# Set up the figure.
fig, tax = ternary.figure(scale=100)
fig.set_size_inches(10, 9)
# Plot the points in groups.
for name, group in df.groupby('Facies Association'):
# Note that we have to shuffle the order.
# This will place Q at the top, F on the left.
# So the column order is: right, top, left.
points = group[['M', 'Q', 'C']].values
tax.scatter(points, marker='o', label=name)
# Plot and label the mean.
tax.scatter([mean], marker='x', s=100)
tax.get_axes().text(29.5, 18.5, "MEAN", color='red')
# Draw some lines.
p1, p2 = (75, 25, 0), (75, 0, 25)
tax.line(p1, p2, linewidth=3, color='k', alpha=0.35, linestyle="-")
p1, p2 = (0, 95, 5), (75, 20, 5)
tax.line(p1, p2, linewidth=3, color='k', alpha=0.35, linestyle="--")
p1, p2 = (15, 0, 85), (15, 85, 0)
tax.line(p1, p2, linewidth=3, color='k', alpha=0.35, linestyle="--")
# Corner labels.
fontsize = 32
offset = 0.1
tax.top_corner_label("Q", fontsize=fontsize, offset=0.2)
tax.left_corner_label("C", fontsize=fontsize, offset=offset)
tax.right_corner_label("M", fontsize=fontsize, offset=offset)
# Decorations.
tax.legend(title="Facies Association")
tax.boundary(linewidth=1)
tax.gridlines(multiple=10, color="gray")
tax.ticks(axis='lbr', linewidth=1, multiple=10)
tax.get_axes().axis('off')
tax.show()
In [15]:
import math
from collections import namedtuple
from typing import NamedTuple
def get_plot_shape(width:float=None, height:float=None) -> NamedTuple:
"""
Helper function to get correctly shaped ternaries.
Get size of plot in inches, given at least one of
width or height. If you give neither, I'll assume
you want a width of 10 inches. If you give both,
I'll use width and ignore the height.
One way to use it is like:
fig, tax = ternary.figure(scale=scale)
fig.set_size_inches(*get_plot_size())
Args:
width (float): Optional width of plot. Defaults
to 10 inches.
height (float): Optional height of plot. Ignored
if width is also provided.
Returns:
NamedTuple: The plot (width, height).
Examples
>>> w, h = get_plot_shape(height=10)
>>> print(f"width: {w:.2f}, height: {h:.2f}")
width: 10.79, height: 10.00
"""
# Derived from theory and experiment...
f = math.sin(math.radians(60)) * 1.07
if (width is None) and (height is None):
width = 10
elif width is None:
width = height / f
if height is None:
height = width * f
shape = namedtuple('Shape', ['width', 'height'])
return shape(width, height)
import doctest
doctest.testmod()
Out[15]:
In [16]:
# Example
w, h = get_plot_shape()
print(f"width: {w:.2f}, height: {h:.2f}")
So that's the kind of ratio we can use to get close to an equilateral triangle.
plotly_express came along in June 2019.
pip install plotly_express
Then you can do this...
In [17]:
import plotly_express as px
px.scatter_ternary(df,
a="Q", b="C", c="M",
color="Facies Association", size="Authigenic",
size_max=15,
)
Not bad for one line of code!
© 2019 Agile Scientific, licensed CC-BY, please share this work