I have a plotly.express.scatter plot with thousands of points. I'd like to add text labels, but only for outliers (eg, far away from a trendline).
How do I do this with plotly?
I'm guessing I need to make a list of points I want labeled and then pass this somehow to plotly (update_layout?). I'm interested in a good way to do this.
Any help appreciated.
You have the right idea: you'll want to have the coordinates of your outliers, and use Plotly's text annotations to add text labels to these points. I am not sure how you want to determine outliers, but the following is an example using the tips dataset.
import pandas as pd
from sklearn import linear_model
import plotly.express as px
df = px.data.tips()
## use linear model to determine outliers by residual
X = df["total_bill"].values.reshape(-1, 1)
y = df["tip"].values
regr = linear_model.LinearRegression()
regr.fit(X, y)
df["predicted_tip"] = regr.predict(X)
df["residual"] = df["tip"] - df["predicted_tip"]
residual_mean, residual_std = df["residual"].mean(), df["residual"].std()
df["residual_normalized"] = (((df["tip"] - df["predicted_tip"]) - residual_mean) / residual_std).abs()
## determine outliers using whatever method you like
outliers = df.loc[df["residual_normalized"] > 3.0, ["total_bill","tip"]]
fig = px.scatter(df, x="total_bill", y="tip", trendline="ols", trendline_color_override="red")
## add text to outliers using their (x,y) coordinates:
for x,y in outliers.itertuples(index=False):
fig.add_annotation(
x=x, y=y,
text="outlier",
showarrow=False,
yshift=10
)
fig.show()
fig.add_annotation. Your answer inspired me to think of another potential solution - making a str column in my data that equals the label value if it's an outlier, or "" otherwise. And then do px.scatter(..., text="that_new_column"). I'll accept your answer and maybe post an additional answer with this other implementation.
px.scatter