Machine Learning Udemy Bootcamp 06 - Seaborn

Seaborn is a statistical plotting library, built on top of matplotlib.
It's designed to work well with pandas dataframe objects.
It's open source and hosted on Github.

conda install seaborn
pip install seaborn

Distribution Plots

Distribution plots allows to show the distribution of a univariate (one variable) set of observations.

DISTPLOT

Distplot is a histogram with a line on it representing the distribution.

import seaborn as sns

# allow visualization in Jupyter Notebook
%matplotlib inline

tips = sns.load_dataset('tips')

tips.head()

# distribution plot
sns.distplot(
    tips['total_bill'],
    # remove kde line and display only histogram
    kde=False,
    # change number of bins
    bins=30
)

JOINTPLOT

Jointplot allows to match up two distplots for bivariate data (two variables).
It creates a canvas where are related two distplots.

sns.jointplot(
    # feature on X
    x='total_bill',
    # feature on Y
    y='tip',
    # dataset where to extract features
    data=tips,
    # hex => hexagonal density plot
    # ref => linear regression visualizer
    # kde => kernel density estimation
    # how data is represented
    kind='scatter'
)

PAIRPLOT

Pairplot allows to plot pairwise relationships across an entire dataframe (for the numerical columns) and supports a color hue argument (for categorical columns).
It's a jointplot for every single combination of the numerical columns in the dataframe.

sns.pairplot(
    # dataset to visualize
    tips,
    # categorical columns
    hue='sex',
    # pass a predefined palette from seaborn website
    palette='coolwarm'
)

RUGPLOT

Rugplot draws a dash mark for every point on a univariate distribution.

sns.rugplot(
    tips['total_bill']
)

KDEPLOT

KDEPlot represents the distribution of data in a KDE (Kernel Density Estimation) format.
Normal distribution is mathematically represented by KDE.

sns.kdeplot(
    tips['total_bill']
)

Categorical Plots

BARPLOT

Barplot is a general plot that allows to aggregate categorical data based off some function, by default the mean.

sns.barplot(
    # feature sex on X (categorical)
    x='sex',
    # feature total_bill on Y (numerical)
    y='total_bill',
    data=tips
)

COUNTPLOT

Countplot is the same as barplot except the estimator is explicitly counting the number of occurrences.

sns.countplot(
    # categorical feature
    x='sex',
    data=tips
)

BOXPLOT

Boxplot shows the distribution of categorical data.
It shows the quartiles of the dataset while the whiskers extend to show the rest of the distribution.
Outliers are plotted as points outside the whiskers.

sns.boxplot(
    # categorical feature
    x='day',
    # numerical feature
    y='total_bill',
    data=tips,
    # categorical feature
    hue='smoker'
)

VIOLINPLOT

Violinplot plays a combination of boxplot and kdeplot.
Allows to understand the relationship between two categorical features and a numerical feature.

sns.violinplot(
    # categorical feature
    x='day',
    # numerical feature
    y='total_bill',
    data=tips,
    hue='sex',
    # split the violin plot by the hue feature
    # instead of having a violin plot for each category of the hue feature
    split=True
)

STRIPPLOT

Stripplot draws a scatterplot where one variable is categorical.
A strip plot can be drawn on its own, but it is also a good complement to a box or violin plot in cases where you want to show all observations along with some representation of the underlying distribution.

sns.stripplot(
    # categorical feature
    x='day',
    # numerical feature
    y='total_bill',
    data=tips,
    # adds a random noise to the data to avoid overlapping
    jitter=True,
    hue='sex'
)

SWARMPLOT

Swarmplot is a combination of stripplot and violinplot, but the points are adjusted (only along the categorical axis) so that they don’t overlap.
This gives a better representation of the distribution of values, although it does not scale as well to large numbers of observations (both in terms of the ability to show all the points and in terms of the computation needed to arrange them).

sns.swarmplot(
    x='day',
    y='total_bill',
    data=tips,
)

FACTORPLOT

Factorplot is the most general form of a categorical plot.
It can take in a kind parameter to adjust the plot type.

sns.factorplot(
    x='day',
    y='total_bill',
    data=tips,
    # specify the kind of plot
    kind='bar'
)

Matrix Plots

Matrix plots allow to plot data as color-encoded matrices and can also be used to indicate clusters within the data.
To be a Marix it should have categorical features on both axes.

HEATMAP

Heatmap is a simple way to plot a matrix plot.

sns.heatmap(
    # dataset to plot
    flights,
    # annotates the heatmap with the numeric value
    annot=True,
    # cmap => colormap
    cmap='coolwarm'
)

CLUSTERMAP

Clustermap uses hierarchal clustering to produce a clustered version of the heatmap.
It will show data aggregated as similar values, heatmap uses the provided order to show the data

sns.clustermap(
    flights,
    # standardize the scale
    standard_scale=1
)

Grids

Grids are general types of plots that allow you to map plot types to rows and columns of a grid, this helps you create similar plots separated by features.

PAIRGRID

Pairgrid is a subplot grid for plotting pairwise relationships in a dataset.
It's how pairplot is implemented, it allows to create a grid of custom plots

Scatterplot:

from matplotlib import pyplot as plt
iris = sns.load_dataset('iris')
g = sns.PairGrid(iris)
# apply scatterplot to the grid
g.map(plt.scatter)

Multiplot:

g = sns.PairGrid(iris)
# apply distplot to diagonal plots
g.map_diag(sns.distplot)
# apply scatterplot to the upper plots
g.map_upper(plt.scatter)
# apply kdeplot to the lower plots
g.map_lower(sns.kdeplot)

FACETGRID

Facetgrid is the general way to create grids of plots based off of a feature.

1 parameter distplot:

tips = sns.load_dataset('tips')
g = sns.FacetGrid(
    # data to use
    data=tips,
    # categorical feature to split the data
    col='time',
    # categorical feature to split the data
    row='smoker'
)
# apply distplot to the grid using the feature total_bill
g.map(sns.distplot, 'total_bill')

2 parameters scatterplot:

g = sns.FacetGrid(
    # data to use
    data=tips,
    # categorical feature to split the data
    col='time',
    # categorical feature to split the data
    row='smoker'
)
# apply distplot to the grid using the feature total_bill
g.map(plt.scatter, 'total_bill', 'tip')

Regression Plots

Regression plots are plots that allow you to create a linear fit between two features.

LMPLOT

import seaborn as sns
tips = sns.load_dataset('tips')
# features separated by hue (color)
sns.lmplot(
    x='total_bill',
    y='tip',
    data=tips,
    hue='sex',
    markers=['o', 'v'],
)

Using different features:

sns.lmplot(
    x='total_bill',
    y='tip',
    data=tips,
    col='day',
    row='time',
    hue='sex',
    aspect=0.6,
)

Styles

Seaborn provides a variety of styles to customize the plots:

import matplotlib.pyplot as plt

# overwrites the default seaborn styles
sns.set_context(
    # paper, notebook, talk, poster
    'poster',
    # font size of the labels
    # font_scale=3
)

# Change the size of the splot using core matplotlib
# It's possible to use matplotlib in combination with seaborn
plt.figure(figsize=(12,3))

sns.set_style(
    # ticks at the edge of the plot
    'ticks'
    # 'darkgrid'
    # 'whitegrid'
)
sns.countplot(x='sex',data=tips)
# remove the top and right spines
sns.despine(top=True, bottom=True)

Using plots parameters is possible to customize the plots even more:

sns.lmplot(
    x='total_bill',
    y='tip',
    data=tips,
    # distribute colors based on a categorical feature
    hue='sex',
    # preset of palettes provided by colormap docs of matplotlib
    palette='seismic'
)