This package contains many statistical recipes for concepts and types introduced in the JuliaStats organization, intended to be used with Plots.jl:

- Types:
- DataFrames (for DataTables support, checkout the
`DataTables`

branch) - Distributions

- DataFrames (for DataTables support, checkout the
- Recipes:
- histogram/histogram2d
- boxplot
- violin
- marginalhist
- corrplot/cornerplot

Initialize:

```
#Pkg.clone("git@github.com:JuliaPlots/StatPlots.jl.git")
using StatPlots
gr(size=(400,300))
```

Table-like data structures, including `DataFrames`

, `IndexedTables`

, `DataStreams`

, etc... (see here for an exhaustive list), are supported thanks to the macro `@df`

which allows passing columns as symbols. Those columns can then be manipulated inside the `plot`

call, like normal `Arrays`

:

```
using DataFrames, IndexedTables
df = DataFrame(a = 1:10, b = 10*rand(10), c = 10 * rand(10))
@df df plot(:a, [:b :c], colour = [:red :blue])
@df df scatter(:a, :b, markersize = 4 * log.(:c + 0.1))
t = IndexedTable(Columns(a = collect(1:10)), Columns(b = rand(10)))
@df t scatter(2 * :b)
```

In case of ambiguity, symbols not referring to `DataFrame`

columns must be escaped by `^()`

:

```
df[:red] = rand(10)
@df df plot(:a, [:b :c], colour = ^([:red :blue]))
```

The `@df`

macro plays nicely with the new syntax of the Query.jl data manipulation package (v0.7 and above), in that a plot command can be added at the end of a query pipeline, without having to explicitly collect the outcome of the query first:

```
using Query, StatPlots
df |>
@where(_.a > 5) |>
@select({_.b, d = _.c-10}) |>
@df scatter(:b, :d)
```

The old syntax, passing the `DataFrame`

as the first argument to the `plot`

call is still supported, but has several limitations (the most important being incompatibility with user recipes):

```
plot(df, :a, [:b :c], colour = [:red :blue])
```

```
using RDatasets
iris = dataset("datasets","iris")
@df iris marginalhist(:PetalLength, :PetalWidth)
```

```
@df iris corrplot([:SepalLength :SepalWidth :PetalLength :PetalWidth], grid = false)
```

or also:

```
@df iris corrplot(cols(1:4), grid = false)
```

A correlation plot may also be produced from a matrix:

```
M = randn(1000,4)
M[:,2] += 0.8sqrt.(abs.(M[:,1])) - 0.5M[:,3] + 5
M[:,3] -= 0.7M[:,1].^2 + 2
corrplot(M, label = ["x$i" for i=1:4])
```

```
cornerplot(M)
```

```
cornerplot(M, compact=true)
```

```
import RDatasets
singers = RDatasets.dataset("lattice","singer")
@df singers violin(:VoicePart,:Height,marker=(0.2,:blue,stroke(0)))
@df singers boxplot!(:VoicePart,:Height,marker=(0.3,:orange,stroke(2)))
```

Asymmetric violin plots can be created using the `side`

keyword (`:both`

- default,`:right`

or `:left`

), e.g.:

```
singers_moscow = deepcopy(singers)
singers_moscow[:Height] = singers_moscow[:Height]+5
@df singers violin(:VoicePart,:Height, side=:right, marker=(0.2,:blue,stroke(0)), label="Scala")
@df singers_moscow violin!(:VoicePart,:Height, side=:left, marker=(0.2,:red,stroke(0)), label="Moscow")
```

The ea-histogram is an alternative histogram implementation, where every 'box' in the histogram contains the same number of sample points and all boxes have the same area. Areas with a higher density of points thus get higher boxes. This type of histogram shows spikes well, but may oversmooth in the tails. The y axis is not intuitively interpretable.

```
a = [randn(100); randn(100)+3; randn(100)/2+3]
ea_histogram(a, bins = :scott, fillalpha = 0.4)
```

```
using Distributions
plot(Normal(3,5), fill=(0, .5,:orange))
```

```
dist = Gamma(2)
scatter(dist, leg=false)
bar!(dist, func=cdf, alpha=0.3)
```

The `qqplot`

function compares the quantiles of two distributions, and accepts either a vector of sample values or a `Distribution`

. The `qqnorm`

is a shorthand for comparing a distribution to the normal distribution. If the distributions are similar the points will be on a straight line.

```
x = rand(Normal(), 100)
y = rand(Cauchy(), 100)
plot(
qqplot(x, y, qqline = :fit), # qqplot of two samples, show a fitted regression line
qqplot(Cauchy, y), # compare with a Cauchy distribution fitted to y; pass an instance (e.g. Normal(0,1)) to compare with a specific distribution
qqnorm(x, qqline = :R) # the :R default line passes through the 1st and 3rd quartiles of the distribution
)
```

```
groupedbar(rand(10,3), bar_position = :stack, bar_width=0.7)
```

This is the default:

```
groupedbar(rand(10,3), bar_position = :dodge, bar_width=0.7)
```

The `group`

syntax is also possible in combination with `groupedbar`

:

```
groupedbar([1, 2, 1, 2, 1, 2], rand(6), group = [1, 1, 2, 2, 3, 3])
```

Population analysis on a table-like data structures can be done using the highly recommended GroupedErrors package.

This external package, in combination with StatPlots, greatly simplifies the creation of two types of plots:

Some simple summary statistics are computed for each experimental subject (mean is default but any scalar valued function would do) and then plotted against some other summary statistics, potentially splitting by some categorical experimental variable.

Some statistical analysis is computed at the single subject level (for example the density/hazard/cumulative of some variable, or the expected value of a variable given another) and the analysis is summarized across subjects (taking for example mean and s.e.m), potentially splitting by some categorical experimental variable.

For more information please refer to the README.

A GUI based on QML and the GR Plots.jl backend to simplify the use of StatPlots.jl and GroupedErrors.jl even further can be found here (usable but still in alpha stage).

07/10/2016

1 day ago

258 commits