Azure Databricks supports various types of visualizations out of the box.

All notebooks support DataFrame visualizations using the display function. The display function also supports rendering image data types and various machine learning visualizations.

Azure Databricks programming language notebooks (Python, Scala, R) support HTML graphics using the displayHTML function; you can pass it any HTML, CSS, or JavaScript code. This function support interactive graphics using JavaScript libraries such as D3.

display function

The easiest way to create a DataFrame visualization in Azure Databricks is to call display(<dataframe-name>). For example, if you have a Spark DataFrame diamonds_df of a diamonds dataset grouped by diamond color, computing the average price, and you call

from pyspark.sql.functions import avg
diamonds_df ="/databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv", header="true", inferSchema="true")


A table of diamond color versus average price displays.



If you see OK with no rendering after calling the display function, mostly likely the DataFrame or collection you passed in is empty.

If you register a Spark DataFrame as a table, you can also query it with SQL to create SQL visualizations.

For Databricks Runtime 5.1 and above, display supports pandas DataFrames.

Visualization types

Click the bar chart icon Chart Button to display a chart of tabular information:


You can click the down arrow next to the bar chart Chart Button to choose another chart type and click Plot Options... to configure the chart.


Images types in DataFrames

display renders columns containing image data types as rich HTML. For clusters running Databricks Runtime 4.1 and above, display attempts to render image thumbnails for DataFrame columns matching Spark’s ImageSchema. Thumbnail rendering works for any images successfully read in through the readImages function. For image values generated through other means, Azure Databricks supports the rendering of 1, 3, or 4 channel images (where each channel consists of a single byte), with the following constraints:

  • One-channel images: mode field must be equal to 0. height, width, and nChannels fields must accurately describe the binary image data in the data field
  • Three-channel images: mode field must be equal to 16. height, width, and nChannels fields must accurately describe the binary image data in the data field. The data field must contain pixel data in three-byte chunks, with the channel ordering (blue, green, red) for each pixel.
  • Four-channel images: mode field must be equal to 24. height, width, and nChannels fields must accurately describe the binary image data in the data field. The data field must contain pixel data in four-byte chunks, with the channel ordering (blue, green, red, alpha) for each pixel.


Suppose you have a folder containing some images:


If you read the images into a DataFrame with ImageSchema.readImages and then display the DataFrame, display renders thumbnails of the images:

from import ImageSchema
image_df = ImageSchema.readImages(sample_img_dir)

Machine learning visualizations

The display function supports various machine learning algorithm visualizations.


For linear and logistic regressions, display supports rendering a fitted versus residuals plot. To obtain this plot, you supply the model and DataFrame.

The following example runs a linear regression on city population to house sale price data and then displays the residuals versus the fitted data.

# Load data
pop_df ="/databricks-datasets/samples/population-vs-price/data_geo.csv", header="true", inferSchema="true")

# Drop rows with missing values and rename the feature and label columns, replacing spaces with _
pop_df = data.dropna() # drop rows with missing values
exprs = [col(column).alias(column.replace(' ', '_')) for column in data.columns]

# Register a UDF to convert the feature (2014_Population_estimate) column vector to a VectorUDT type and apply it to the column.
from import Vectors, VectorUDT

spark.udf.register("oneElementVec", lambda d: Vectors.dense([d]), returnType=VectorUDT())
tdata =*exprs).selectExpr("oneElementVec(2014_Population_estimate) as features", "2015_median_sales_price as label")

# Run a linear regression
from import LinearRegression

lr = LinearRegression()
modelA =, {lr.regParam:0.0})

# Plot residuals versus fitted data
display(modelA, tdata)

ROC curves

For logistic regressions, display supports rendering an ROC curve. To obtain this plot, you supply the model, the prepped data that is input to the fit method, and the parameter "ROC".

The following example develops a classifier that predicts if an individual earns <=50K or >50k a year from various attributes of the individual. The Adult dataset derives from census data, and consists of information about 48842 individuals and their annual income.

  age DOUBLE,
  workclass STRING,
  fnlwgt DOUBLE,
  education STRING,
  education_num DOUBLE,
  marital_status STRING,
  occupation STRING,
  relationship STRING,
  race STRING,
  sex STRING,
  capital_gain DOUBLE,
  capital_loss DOUBLE,
  hours_per_week DOUBLE,
  native_country STRING,
  income STRING)
OPTIONS (path "/databricks-datasets/adult/", header "true")
dataset = spark.table("adult")

# Use One-Hot Encoding to convert all categorical variables into binary vectors.

from import Pipeline
from import OneHotEncoderEstimator, StringIndexer, VectorAssembler
categoricalColumns = ["workclass", "education", "marital_status", "occupation", "relationship", "race", "sex", "native_country"]

stages = [] # stages in our Pipeline
for categoricalCol in categoricalColumns:
    # Category Indexing with StringIndexer
    stringIndexer = StringIndexer(inputCol=categoricalCol, outputCol=categoricalCol + "Index")
    # Use OneHotEncoder to convert categorical variables into binary SparseVectors
    # encoder = OneHotEncoderEstimator(inputCol=categoricalCol + "Index", outputCol=categoricalCol + "classVec")
    encoder = OneHotEncoderEstimator(inputCols=[stringIndexer.getOutputCol()], outputCols=[categoricalCol + "classVec"])
    # Add stages.  These are not run here, but will run all at once later on.
    stages += [stringIndexer, encoder]

# Convert label into label indices using the StringIndexer
label_stringIdx = StringIndexer(inputCol="income", outputCol="label")
stages += [label_stringIdx]

# Transform all features into a vector using VectorAssembler
numericCols = ["age", "fnlwgt", "education_num", "capital_gain", "capital_loss", "hours_per_week"]
assemblerInputs = [c + "classVec" for c in categoricalColumns] + numericCols
assembler = VectorAssembler(inputCols=assemblerInputs, outputCol="features")
stages += [assembler]

# Run the stages as a Pipeline. This puts the data through all of the feature transformations in a single call.

partialPipeline = Pipeline().setStages(stages)
pipelineModel =
preppedDataDF = pipelineModel.transform(dataset)

# Fit logistic regression model

from import LogisticRegression
lrModel = LogisticRegression().fit(preppedDataDF)

# ROC for data
display(lrModel, preppedDataDF, "ROC")

To display the residuals, omit the "ROC" parameter:

display(lrModel, preppedDataDF)

Decision trees

The display function supports rendering a decision tree. This is supported for Scala in Databricks Runtime 4.1 and above and for Python in Databricks Runtime 4.3 and above.

To obtain this visualization, you supply the decision tree model.

The following examples train a tree to recognize digits (0 - 9) from the MNIST dataset of images of handwritten digits and then displays the tree.


val trainingDF ="libsvm").load("/databricks-datasets/mnist-digits/data-001/mnist-digits-train.txt").cache
val testDF ="libsvm").load("/databricks-datasets/mnist-digits/data-001/mnist-digits-test.txt").cache

import{DecisionTreeClassifier, DecisionTreeClassificationModel}

val indexer = new StringIndexer().setInputCol("label").setOutputCol("indexedLabel")
val dtc = new DecisionTreeClassifier().setLabelCol("indexedLabel")
val pipeline = new Pipeline().setStages(Array(indexer, dtc))

val model =
val tree = model.stages.last.asInstanceOf[DecisionTreeClassificationModel]



trainingDF ="libsvm").load("/databricks-datasets/mnist-digits/data-001/mnist-digits-train.txt").cache()
testDF ="libsvm").load("/databricks-datasets/mnist-digits/data-001/mnist-digits-test.txt").cache()

from import DecisionTreeClassifier
from import StringIndexer
from import Pipeline

indexer = StringIndexer().setInputCol("label").setOutputCol("indexedLabel")

dtc = DecisionTreeClassifier().setLabelCol("indexedLabel")

# Chain indexer + dtc together into a single ML Pipeline.
pipeline = Pipeline().setStages([indexer, dtc])

model =

Visualizations in Python

To create a visualization in Python, call display(<dataframe-name>).

diamonds_df ="/databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv", header="true", inferSchema="true")


For a deep dive into Python visualizations using display, see Visualization Deep Dive in Python.

For visualizations specific to machine learning, see Machine learning visualizations.

You can also use other Python libraries to generate visualizations. The Databricks Runtime includes the seaborn visualization library so it’s easy to create a seaborn plot. For example:

import seaborn as sns

df = sns.load_dataset("iris")
g = sns.PairGrid(df, diag_sharey=False)
g.map_diag(sns.kdeplot, lw=3)



For other libraries and examples, see Matplotlib and ggplot in Python Notebooks, Bokeh in Python Notebooks, and Plotly in Python and R Notebooks.

Visualizations in R

In addition to the Azure Databricks visualizations, R notebooks can use any R visualization package. The R notebook will capture the resulting plot as a .png and display it inline.

display function

diamonds_df <- read.df("/databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv", source = "csv", header="true", inferSchema = "true")

display(arrange(agg(groupBy(diamonds_df, "color"), "price" = "avg"), "color"))

Default library

fit <- lm(Petal.Length ~., data = iris)
layout(matrix(c(1,2,3,4),2,2)) # optional 4 graphs/page


ggplot(diamonds, aes(carat, price, color = color, group = 1)) + geom_point(alpha = 0.3) + stat_smooth()


xyplot(price ~ carat | cut, diamonds, scales = list(log = TRUE), type = c("p", "g", "smooth"), ylab = "Log price")

You can also install and use other plotting libraries.

install.packages("DandEFA", repos = "")
timss2011 <- na.omit(timss2011)
dandpal <- rev(rainbow(100, start = 0, end = 0.2))
facl <- factload(timss2011,nfac=5,method="prax",cormeth="spearman")
facl <- factload(timss2011,nfac=8,method="mle",cormeth="pearson")

Also see htmlwidgets in R Notebooks and Plotly in Python and R Notebooks.

Visualizations in Scala

The easiest way to perform plotting in Scala is to use the display method. For example:

val diamonds_df ="csv").option("header","true").option("inferSchema","true").load("/databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv")


For a deep dive into Scala visualizations using display, see Visualization Deep Dive in Scala.

For visualizations specific to machine learning, see Machine learning visualizations.

Visualizations in SQL

When you execute SQL, Azure Databricks automatically extracts some of the data and displays it as a table.

SELECT color, avg(price) AS price FROM diamonds GROUP BY color ORDER BY COLOR

From there you can select different styles of visualization.


displayHTML function

The displayHTML iframe is served from the domain, and the iframe sandbox includes the allow-same-origin attribute. must be accessible from your browser. If it is currently blocked by your corporate network, it will need to be whitelisted by IT.

See Embed static images in notebooks and HTML, D3, and SVG in Notebooks for examples of using displayHTML.