Implementation of "value_counts()" like feature in bar chart plotly express

[matsujju]

Will it be possible to add count feature/frequency as axis label for bar chart like in seaborn for plotly express in future ?
It is helpful when we need like specific number of values from particular feature while plotting.

[nicolaskruchten]

This this basically what px.histogram() does, and it also supports other functions like sum etc... Does that meet your needs or am I misunderstanding?

[matsujju]

I am not sure how px.histogram() uses value_counts() for calculating frequencies of let say xaxis unique categories and plotting them in yaxis accordingly like in seaborn. As on looking the parameters, it requires both x and y to plot. However, if i have to let say plot top10 uniques categories with it's frequency, how to do that with histogram? Do i need to calculate frequency of top10 categories and pass it to yaxis, if this the case then this will have to be done in all other parameters which are directly affected by serialization of values in pandas dataframe column?

[nicolaskruchten]

if you pass in just x then px.histogram() will bin along that dimension and summarize by count. If you pass in x and y, it will bin along x and summarize by sum(y) unless orientation="h" in which case it'll do the same thing the other way. If you pass in just y it will infer orientation="h" and bin along y, summarizing by count also.

[nicolaskruchten]

Simple example:

[matsujju]

thanks...it is much helpful though i didn't find anywhere related to this in documentation( or may be i just missed out). However, there is one issue, how to consider only let say in your above example top2 categories only i.e days ?

[nicolaskruchten]

Right, at the moment there's no built-in "top N" support. How does Seaborn handle this?

[matsujju]

Sorry but i can't find any parameter built in px.histogram(). Can you please specify what are you implying?
Seaborn has two plots for this, one is countplot with order parameter where we can specify the index of top categories using pandas value_counts() method and same with catplot with kind = 'count' and order parameters.

[nicolaskruchten]

Sorry, typo fixed: "there's no built in feature"

[matsujju]

Can you suggest how do I implement this in plotly express? and will this feature be added in future?

[nicolaskruchten]

I don't think this is the kind of feature we'd add to the library at this time, as this is more of a data-prep concern than a strictly visualization concern: Plotly Express visualizes the whole dataset you provide to it, as a rule, rather than a subset :)

That said, what you're looking for looks a little like this?

import plotly.express as px
df = px.data.tips()
px.bar(df["day"].value_counts()[:2])