Ten things you can do on the Windows Data Science Virtual Machine

The Windows Data Science Virtual Machine (DSVM) is a powerful data science development environment where you can perform data exploration and modeling tasks. The environment comes already built and bundled with several popular data analytics tools that make it easy to get started with your analysis for on-premises, cloud, or hybrid deployments.

The DSVM works closely with Azure services. It can read and process data that's already stored on Azure, in Azure SQL Data Warehouse, Azure Data Lake, Azure Storage, or Azure Cosmos DB. It can also take advantage of other analytics tools, such as Azure Machine Learning and Azure Data Factory.

In this article, you'll learn how to use your DSVM to perform data science tasks and interact with other Azure services. Here are some of the things you can do on the DSVM:

  • Explore data and develop models locally on the DSVM by using Microsoft Machine Learning Server and Python.
  • Use a Jupyter notebook to experiment with your data in a browser by using Python 2, Python 3, and Microsoft R. (Microsoft R is an enterprise-ready version of R designed for performance.)
  • Deploy models built through R and Python on Azure Machine Learning so client applications can access your models by using a simple web service interface.
  • Administer your Azure resources by using the Azure portal or PowerShell.
  • Extend your storage space and share large-scale datasets/code across your whole team by creating an Azure Files share as a mountable drive on your DSVM.
  • Share code with your team by using GitHub. Access your repository by using the pre-installed Git clients: Git Bash and Git GUI.
  • Access Azure data and analytics services like Azure Blob storage, Azure Data Lake, Azure HDInsight (Hadoop), Azure Cosmos DB, Azure SQL Data Warehouse, and Azure SQL Database.
  • Build reports and a dashboard by using the Power BI Desktop instance that's pre-installed on the DSVM, and deploy them in the cloud.
  • Dynamically scale your DSVM to meet your project's needs.
  • Install additional tools on your virtual machine.


Additional usage charges apply for many of the data storage and analytics services listed in this article. For details, see the Azure pricing page.



This article has been updated to use the new Azure PowerShell Az module. You can still use the AzureRM module, which will continue to receive bug fixes until at least December 2020. To learn more about the new Az module and AzureRM compatibility, see Introducing the new Azure PowerShell Az module. For Az module installation instructions, see Install Azure PowerShell.

Explore data and develop models with Microsoft Machine Learning Server

You can use languages like R and Python to do your data analytics right on the DSVM.

For R, you can use an IDE like RStudio that can be found on the start menu or on the desktop. Or you can use R Tools for Visual Studio. Microsoft has provided additional libraries on top of the open-source CRAN R to enable scalable analytics and the ability to analyze data larger than the memory size allowed in parallel chunked analysis.

For Python, you can use an IDE like Visual Studio Community Edition, which has the Python Tools for Visual Studio (PTVS) extension pre-installed. By default, only Python 3.6, the root Conda environment, is configured on PTVS. To enable Anaconda Python 2.7, take the following steps:

  1. Create custom environments for each version by going to Tools > Python Tools > Python Environments, and then selecting + Custom in Visual Studio Community Edition.
  2. Give a description and set the environment prefix path as c:\anaconda\envs\python2 for Anaconda Python 2.7.
  3. Select Auto Detect > Apply to save the environment.

See the PTVS documentation for more details on how to create Python environments.

Now you're set up to create a new Python project. Go to File > New > Project > Python and select the type of Python application you're building. You can set the Python environment for the current project to the desired version (Python 2.7 or 3.6) by right-clicking Python environments and then selecting Add/Remove Python Environments. You can find more information about working with PTVS in the product documentation.

Use Jupyter notebooks

The Jupyter Notebook provides a browser-based IDE for data exploration and modeling. You can use Python 2, Python 3, or R (both open source and Microsoft R Server) in a Jupyter notebook.

To start the Jupyter Notebook, select the Jupyter Notebook icon on the Start menu or on the desktop. In the DSVM command prompt, you can also run the command jupyter notebook from the directory where you have existing notebooks or where you want to create new notebooks.

After you start Jupyter, you should see a directory that contains a few example notebooks that are pre-packaged into the DSVM. Now you can:

  • Select the notebook to see the code.
  • Run each cell by selecting Shift+Enter.
  • Run the entire notebook by selecting Cell > Run.
  • Create a new notebook by selecting the Jupyter icon (upper-left corner), selecting the New button on the right, and then choosing the notebook language (also known as kernels).


Currently, Python 2.7, Python 3.6, R, Julia, and PySpark kernels in Jupyter are supported. The R kernel supports programming in both open-source R and Microsoft R.

When you're in the notebook, you can explore your data, build the model, and test the model by using your choice of libraries.

Train and deploy models by using Azure Machine Learning

After you've built and validated your model, the next step is usually to deploy it into production. This step allows your client applications to invoke the model predictions on a real-time basis or a batch mode basis. Azure Machine Learning provides a mechanism to operationalize a model built in either R or Python.

When you operationalize your model in Azure Machine Learning, a web service is exposed. It allows clients to make REST calls that pass in input parameters and receive predictions from the model as outputs.

Build and operationalize Python models

Here's a snippet of code developed in a Python Jupyter notebook that builds a simple model by using the Scikit-learn library:

# IRIS classification
from sklearn import datasets
from sklearn import svm
clf = svm.SVC()
iris = datasets.load_iris()
X, y = iris.data, iris.target
clf.fit(X, y)

The method that's used to deploy your Python models to Azure Machine Learning wraps the prediction of the model into a function and decorates it with attributes provided by the pre-installed Azure Machine Learning Python library. The attributes denote your Azure Machine Learning workspace ID, the API key, and the input and return parameters.

from azureml import services
@services.publish(workspaceid, auth_token)
@services.types(sep_l=float, sep_w=float, pet_l=float, pet_w=float)
@services.returns(int)  # 0, or 1, or 2
def predictIris(sep_l, sep_w, pet_l, pet_w):
    inputArray = [sep_l, sep_w, pet_l, pet_w]

return clf.predict(inputArray)

A client can now make calls to the web service. Convenience wrappers construct the REST API requests. Here's sample code to consume the web service:

# Consume through web service URL and keys
from azureml import services
@services.service(url, api_key)
@services.types(sep_l = float, sep_w = float, pet_l=float, pet_w=float)
def IrisPredictor(sep_l, sep_w, pet_l, pet_w):



Currently, the Azure Machine Learning library is supported only on Python 2.7.

Build and operationalize R models

You can deploy R models built on the Data Science Virtual Machine or elsewhere onto Azure Machine Learning in a way that's similar to how it's done for Python. Here are the steps:

  1. Create a settings.json file to provide your workspace ID and authentication token.
  2. Write a wrapper for the model's predict function.
  3. Call publishWebService in the Azure Machine Learning library to pass in the function wrapper.

Use the following procedure and code snippets to set up, build, publish, and consume a model as a web service in Azure Machine Learning.

Set up

Create a settings.json file under a directory called .azureml under your home directory. Enter the parameters from your Azure Machine Learning workspace.

Here's the settings.json file structure:

"id"                  : "ENTER YOUR AZUREML WORKSPACE ID",
"authorization_token" : "ENTER YOUR AZUREML AUTH TOKEN"

Build a model in R and publish it in Azure Machine Learning

ws <- workspace(config="~/.azureml/settings.json")

if(!require("lme4")) install.packages("lme4")
train <- sleepstudy[sample(nrow(sleepstudy), 120),]
m <- lm(Reaction ~ Days + Subject, data = train)

# Define a prediction function to publish based on the model:
sleepyPredict <- function(newdata){
        predict(m, newdata=newdata)

ep <- publishWebService(ws, fun = sleepyPredict, name="sleepy lm", inputSchema = sleepstudy, data.frame=TRUE)

Consume the model deployed in Azure Machine Learning

To consume the model from a client application, use the Azure Machine Learning library to look up the published web service by name. Use the services API call to determine the endpoint. Then you just call the consume function and pass in the data frame to be predicted.

Use the following code to consume the model published as an Azure Machine Learning web service:

ws <- workspace(config="~/.azureml/settings.json")

s <-  services(ws, name = "sleepy lm")
s <- tail(s, 1) # use the last published function, in case of duplicate function names

ep <- endpoints(ws, s)

# Try this out, and compare with raw data
ans = consume(ep, sleepstudy)$ans

See more information about R packages in Machine Learning Studio.

Manage Azure resources

The DSVM doesn't just allow you to build your analytics solution locally on the virtual machine. It also allows you to access services on the Azure cloud platform. Azure provides several compute, storage, data analytics, and other services that you can administer and access from your DSVM.

To administer your Azure subscription and cloud resources, you have two options:

Extend storage by using shared file systems

Data scientists can share large datasets, code, or other resources within the team. The DSVM has about 45 GB of space available. To extend your storage, you can use Azure Files and either mount it on one or more DSVM instances or access it via a REST API. You can also use the Azure portal or use Azure PowerShell to add extra dedicated data disks.


The maximum space on the Azure Files share is 5 TB. The size limit for each file is 1 TB.

You can use this script in Azure PowerShell to create an Azure Files share:

# Authenticate to Azure.
# Select your subscription
Get-AzSubscription –SubscriptionName "<your subscription name>" | Select-AzSubscription
# Create a new resource group.
New-AzResourceGroup -Name <dsvmdatarg>
# Create a new storage account. You can reuse existing storage account if you want.
New-AzStorageAccount -Name <mydatadisk> -ResourceGroupName <dsvmdatarg> -Location "<Azure Data Center Name For eg. South Central US>" -Type "Standard_LRS"
# Set your current working storage account
Set-AzCurrentStorageAccount –ResourceGroupName "<dsvmdatarg>" –StorageAccountName <mydatadisk>

# Create an Azure Files share
$s = New-AzStorageShare <<teamsharename>>
# Create a directory under the file share. You can give it any name
New-AzStorageDirectory -Share $s -Path <directory name>
# List the share to confirm that everything worked
Get-AzStorageFile -Share $s

Now that you have created an Azure Files share, you can mount it in any virtual machine in Azure. We recommend that you put the VM in the same Azure datacenter as the storage account, to avoid latency and data transfer charges. Here are the Azure PowerShell commands to mount the drive on the DSVM:

# Get the storage key of the storage account that has the Azure Files share from the Azure portal. Store it securely on the VM to avoid being prompted in the next command.
cmdkey /add:<<mydatadisk>>.file.core.windows.net /user:<<mydatadisk>> /pass:<storage key>

# Mount the Azure Files share as drive Z on the VM. You can choose another drive letter if you want.
net use z:  \\<mydatadisk>.file.core.windows.net\<<teamsharename>>

Now you can access this drive as you would any normal drive on the VM.

Share code in GitHub

GitHub is a code repository where you can find code samples and sources for various tools by using technologies shared by the developer community. It uses Git as the technology to track and store versions of the code files. GitHub is also a platform where you can create your own repository to store your team's shared code and documentation, implement version control, and control who has access to view and contribute code.

Visit the GitHub help pages for more information on using Git. You can use GitHub as one of the ways to collaborate with your team, use code developed by the community, and contribute code back to the community.

The DSVM comes loaded with client tools on the command line and on the GUI to access the GitHub repository. The command-line tool that works with Git and GitHub is called Git Bash. Visual Studio is installed on the DSVM and has the Git extensions. You can find icons for these tools on the Start menu and on the desktop.

To download code from a GitHub repository, you use the git clone command. For example, to download the data science repository published by Microsoft into the current directory, you can run the following command in Git Bash:

git clone https://github.com/Azure/DataScienceVM.git

In Visual Studio, you can do the same clone operation. The following screenshot shows how to access Git and GitHub tools in Visual Studio:

Screenshot of Visual Studio with the GitHub connection displayed

You can find more information on using Git to work with your GitHub repository from resources available on github.com. The cheat sheet is a useful reference.

Access Azure data and analytics services

Azure Blob storage

Azure Blob storage is a reliable, economical cloud storage service for data big and small. This section describes how you can move data to Blob storage and access data stored in an Azure blob.


  • Create your Azure Blob storage account from the Azure portal.

    Screenshot of the storage account creation process in the Azure portal

  • Confirm that the command-line AzCopy tool is pre-installed: C:\Program Files (x86)\Microsoft SDKs\Azure\AzCopy\azcopy.exe. The directory that contains azcopy.exe is already on your PATH environment variable, so you can avoid typing the full command path when running this tool. For more information on the AzCopy tool, see the AzCopy documentation.

  • Start the Azure Storage Explorer tool. You can download it from the Storage Explorer webpage.

    Screenshot of Azure Storage Explorer accessing a storage account

Move data from a VM to an Azure blob: AzCopy

To move data between your local files and Blob storage, you can use AzCopy on the command line or in PowerShell:

AzCopy /Source:C:\myfolder /Dest:https://<mystorageaccount>.blob.core.windows.net/<mycontainer> /DestKey:<storage account key> /Pattern:abc.txt

Replace C:\myfolder with the path where your file is stored, mystorageaccount with your Blob storage account name, mycontainer with the container name, and storage account key with your Blob storage access key. You can find your storage account credentials in the Azure portal.

Run the AzCopy command in PowerShell or from a command prompt. Here is some example usage of the AzCopy command:

# Copy *.sql from a local machine to an Azure blob
"C:\Program Files (x86)\Microsoft SDKs\Azure\AzCopy\azcopy" /Source:"c:\Aaqs\Data Science Scripts" /Dest:https://[ENTER STORAGE ACCOUNT].blob.core.windows.net/[ENTER CONTAINER] /DestKey:[ENTER STORAGE KEY] /S /Pattern:*.sql

# Copy back all files from an Azure blob container to a local machine

"C:\Program Files (x86)\Microsoft SDKs\Azure\AzCopy\azcopy" /Dest:"c:\Aaqs\Data Science Scripts\temp" /Source:https://[ENTER STORAGE ACCOUNT].blob.core.windows.net/[ENTER CONTAINER] /SourceKey:[ENTER STORAGE KEY] /S

After you run the AzCopy command to copy to an Azure blob, your file will appear in Azure Storage Explorer.

Screenshot of the storage account, displaying the uploaded CSV file

Move data from a VM to an Azure blob: Azure Storage Explorer

You can also upload data from the local file in your VM by using Azure Storage Explorer:

  • To upload data to a container, select the target container and select the Upload button.Screenshot of the upload button in Azure Storage Explorer
  • Select the ellipsis (...) to the right of the Files box, select one or multiple files to upload from the file system, and select Upload to begin uploading the files.Screenshot of the Upload files dialog box

Read data from an Azure blob: Machine Learning reader module

In Azure Machine Learning Studio, you can use the Import Data module to read data from your blob.

Screenshot of the Import Data module in Machine Learning Studio

Read data from an Azure blob: Python ODBC

You can use the BlobService library to read data directly from a blob in a Jupyter notebook or in a Python program.

First, import the required packages:

import pandas as pd
from pandas import Series, DataFrame
import numpy as np
import matplotlib.pyplot as plt
from time import time
import pyodbc
import os
from azure.storage.blob import BlobService
import tables
import time
import zipfile
import random

Then, plug in your Blob storage account credentials and read data from the blob:

STORAGEACCOUNTKEY = 'xxxxxxxxxxxxxxxx'
BLOBNAME = 'nyctaxidataset/nyctaxitrip/trip_data_1.csv'
localfilename = 'trip_data_1.csv'
LOCALDIRECTORY = os.getcwd()
LOCALFILE =  os.path.join(LOCALDIRECTORY, localfilename)

#download from blob
t1 = time.time()
blob_service = BlobService(account_name=STORAGEACCOUNTNAME,account_key=STORAGEACCOUNTKEY)
t2 = time.time()
print(("It takes %s seconds to download "+BLOBNAME) % (t2 - t1))

#unzip downloaded files if needed
#with zipfile.ZipFile(ZIPPEDLOCALFILE, "r") as z:
#    z.extractall(LOCALDIRECTORY)

df1 = pd.read_csv(LOCALFILE, header=0)
df1.columns = ['medallion','hack_license','vendor_id','rate_code','store_and_fwd_flag','pickup_datetime','dropoff_datetime','passenger_count','trip_time_in_secs','trip_distance','pickup_longitude','pickup_latitude','dropoff_longitude','dropoff_latitude']
print 'the size of the data is: %d rows and  %d columns' % df1.shape

The data is read as a data frame:

Screenshot of the first 10 rows of data

Azure Data Lake

Azure Data Lake Storage is a hyperscale repository for big data analytics workloads and is compatible with Hadoop Distributed File System (HDFS). It works with Hadoop, Spark, and Azure Data Lake Analytics. In this section, you'll learn how you can move data into Azure Data Lake Storage and run analytics by using Azure Data Lake Analytics.


Move data from a VM to Data Lake: Azure Data Lake Explorer

You can use Azure Data Lake Explorer to upload data from the local files in your virtual machine to Data Lake Storage.

You can also build a data pipeline to operationalize your data movement to or from Azure Data Lake by using Azure Data Factory. This article guides you through the steps to build the data pipelines.

Read data from an Azure blob to Data Lake: U-SQL

If your data resides in Azure Blob storage, you can directly read data from an Azure blob in a U-SQL query. Before you compose your U-SQL query, make sure your Blob storage account is linked to your Azure Data Lake instance. Go to the Azure portal, find your Azure Data Lake Analytics dashboard, select Add Data Source, select a storage type of Azure Storage, and plug in your Azure storage account name and key. Then you can reference the data stored in the storage account.

Screenshot of the Add Data Source dialog box

In Visual Studio, you can read data from Blob storage, manipulate data, engineer features, and send the resulting data to either Azure Data Lake or Azure Blob storage. When you reference the data in Blob storage, use wasb://. When you reference the data in Azure Data Lake, use swbhdfs://.

You can use the following U-SQL queries in Visual Studio:

@a =
    EXTRACT medallion string,
            hack_license string,
            vendor_id string,
            rate_code string,
            store_and_fwd_flag string,
            pickup_datetime string,
            dropoff_datetime string,
            passenger_count int,
            trip_time_in_secs double,
            trip_distance double,
            pickup_longitude string,
            pickup_latitude string,
            dropoff_longitude string,
            dropoff_latitude string

    FROM "wasb://<Container name>@<Azure Blob Storage Account Name>.blob.core.windows.net/<Input Data File Name>"
    USING Extractors.Csv();

@b =
    SELECT vendor_id,
    COUNT(medallion) AS cnt_medallion,
    SUM(passenger_count) AS cnt_passenger,
    AVG(trip_distance) AS avg_trip_dist,
    MIN(trip_distance) AS min_trip_dist,
    MAX(trip_distance) AS max_trip_dist,
    AVG(trip_time_in_secs) AS avg_trip_time
    FROM @a
    GROUP BY vendor_id;

OUTPUT @b   
TO "swebhdfs://<Azure Data Lake Storage Account Name>.azuredatalakestore.net/<Folder Name>/<Output Data File Name>"
USING Outputters.Csv();

OUTPUT @b   
TO "wasb://<Container name>@<Azure Blob Storage Account Name>.blob.core.windows.net/<Output Data File Name>"
USING Outputters.Csv();

After your query is submitted to the server, a diagram shows the status of your job.

Screenshot of the job status diagram

Query data in Data Lake: U-SQL

After the dataset is ingested in Azure Data Lake, you can use U-SQL language to query and explore the data. U-SQL language is similar to T-SQL, but combines some features from C# so that users can write customized modules and user-defined functions. You can use the scripts in the previous step.

After the query is submitted to the server, tripdata_summary.CSV appears in Azure Data Lake Explorer. You can preview the data by right-clicking the file.

Screenshot of the CSV file in Data Lake Explorer

The file information appears:

Screenshot of the file summary information

HDInsight Hadoop clusters

Azure HDInsight is a managed Apache Hadoop, Spark, HBase, and Storm service in the cloud. You can work easily with Azure HDInsight clusters from the Data Science Virtual Machine.


  • Create your Azure Blob storage account from the Azure portal. This storage account is used to store data for HDInsight clusters.

    Screenshot of creating a storage account from the Azure portal

  • Customize Azure HDInsight Hadoop clusters from the Azure portal.

    Link the storage account created with your HDInsight cluster when it's created. This storage account is used for accessing data that can be processed within the cluster.

    Selections for linking the storage account created with an HDInsight cluster

  • Enable Remote Desktop access to the head node of the cluster after it's created. Remember the remote access credentials that you specify here, because you'll need them in the subsequent procedure.

    Remote Desktop button for enabling remote access to the HDInsight cluster

  • Create an Azure Machine Learning workspace. Your Machine Learning experiments are stored in this Machine Learning workspace. Select the highlighted options in the portal, as shown in the following screenshot:

    Create an Azure Machine Learning workspace

  • Enter the parameters for your workspace.

    Enter Machine Learning workspace parameters

  • Upload data by using IPython Notebook. Import required packages, plug in credentials, create a database in your storage account, and then load data into HDI clusters.

# Import required packages
import pyodbc
import time as time
import json
import os
import urllib
import urllib2
import warnings
import re
import pandas as pd
import matplotlib.pyplot as plt
from azure.storage.blob import BlobService
warnings.filterwarnings("ignore", category=UserWarning, module='urllib2')

# Create the connection to Hive by using ODBC
SERVER_NAME = 'xxx.azurehdinsight.net'
DATABASE_NAME = 'nyctaxidb'
USERID = 'xxx'
PASSWORD = 'xxxx'
DB_DRIVER = 'Microsoft Hive ODBC Driver'
driver = 'DRIVER={' + DB_DRIVER + '}'
server = 'Host=' + SERVER_NAME + ';Port=443'
database = 'Schema=' + DATABASE_NAME
hiveserv = 'HiveServerType=2'
auth = 'AuthMech=6'
uid = 'UID=' + USERID
pwd = 'PWD=' + PASSWORD
    [driver, server, database, hiveserv, auth, uid, pwd])
connection = pyodbc.connect(CONNECTION_STRING, autocommit=True)
cursor = connection.cursor()

# Create the Hive database and tables
queryString = "create database if not exists nyctaxidb;"

queryString = """
                create external table if not exists nyctaxidb.trip
                    medallion string,
                    hack_license string,
                    vendor_id string,
                    rate_code string,
                    store_and_fwd_flag string,
                    pickup_datetime string,
                    dropoff_datetime string,
                    passenger_count int,
                    trip_time_in_secs double,
                    trip_distance double,
                    pickup_longitude double,
                    pickup_latitude double,
                    dropoff_longitude double,
                    dropoff_latitude double)  
                PARTITIONED BY (month int)
                ROW FORMAT DELIMITED FIELDS TERMINATED BY ',' lines terminated by '\\n'
                STORED AS TEXTFILE LOCATION 'wasb:///nyctaxidbdata/trip' TBLPROPERTIES('skip.header.line.count'='1');

queryString = """
                create external table if not exists nyctaxidb.fare
                    medallion string,
                    hack_license string,
                    vendor_id string,
                    pickup_datetime string,
                    payment_type string,
                    fare_amount double,
                    surcharge double,
                    mta_tax double,
                    tip_amount double,
                    tolls_amount double,
                    total_amount double)
                PARTITIONED BY (month int)
                ROW FORMAT DELIMITED FIELDS TERMINATED BY ',' lines terminated by '\\n'
                STORED AS TEXTFILE LOCATION 'wasb:///nyctaxidbdata/fare' TBLPROPERTIES('skip.header.line.count'='1');

# Upload data from Blob storage to an HDI cluster
for i in range(1, 13):
    queryString = "LOAD DATA INPATH 'wasb:///nyctaxitripraw2/trip_data_%d.csv' INTO TABLE nyctaxidb2.trip PARTITION (month=%d);" % (
        i, i)
    queryString = "LOAD DATA INPATH 'wasb:///nyctaxifareraw2/trip_fare_%d.csv' INTO TABLE nyctaxidb2.fare PARTITION (month=%d);" % (
        i, i)

Alternatively, you can follow this walkthrough to upload NYC Taxi data to the HDI cluster. Major steps include:

  • Use AzCopy to download zipped CSVs from the public blob to your local folder.
  • Use AzCopy to upload unzipped CSVs from the local folder to an HDI cluster.
  • Log in to the head node of Hadoop cluster and prepare for exploratory data analysis.

After the data is loaded into the HDI cluster, you can check your data in Azure Storage Explorer. And the nyctaxidb database has been created in the HDI cluster.

Data exploration: Hive Queries in Python

Because the data is in a Hadoop cluster, you can use the pyodbc package to connect to Hadoop clusters and query databases by using Hive to do exploration and feature engineering. You can view the existing tables that you created in the prerequisite step.

queryString = """
    show tables in nyctaxidb2;
pd.read_sql(queryString, connection)

View existing tables

Let's look at the number of records in each month and the frequencies of tipped or not in the trip table:

queryString = """
    select month, count(*) from nyctaxidb.trip group by month;
results = pd.read_sql(queryString,connection)

%matplotlib inline

results.columns = ['month', 'trip_count']
df = results.copy()
df.index = df['month']

Plot of number of records in each month

queryString = """
    SELECT tipped, COUNT(*) AS tip_freq
        SELECT if(tip_amount > 0, 1, 0) as tipped, tip_amount
        FROM nyctaxidb.fare
    GROUP BY tipped;
results = pd.read_sql(queryString, connection)

results.columns = ['tipped', 'trip_count']
df = results.copy()
df.index = df['tipped']

Plot of tip frequencies

You can also compute the distance between pickup location and drop-off location, and then compare it to the trip distance.

queryString = """
                select pickup_longitude, pickup_latitude, dropoff_longitude, dropoff_latitude, trip_distance, trip_time_in_secs,
                    pow(sin((dropoff_longitude-pickup_longitude)*radians(180)/180/2),2))) as direct_distance
                    from nyctaxidb.trip
                    where month=1
                        and pickup_longitude between -90 and -30
                        and pickup_latitude between 30 and 90
                        and dropoff_longitude between -90 and -30
                        and dropoff_latitude between 30 and 90;
results = pd.read_sql(queryString, connection)

Top rows of the pickup and drop-off table

results.columns = ['pickup_longitude', 'pickup_latitude', 'dropoff_longitude',
                   'dropoff_latitude', 'trip_distance', 'trip_time_in_secs', 'direct_distance']
df = results.loc[results['trip_distance'] <= 100]  # remove outliers
df = df.loc[df['direct_distance'] <= 100]  # remove outliers
plt.scatter(df['direct_distance'], df['trip_distance'])

Plot of pickup/drop-off distance to trip distance

Now let's prepare a downsampled (1 percent) set of data for modeling. You can use this data in the Machine Learning reader module.

queryString = """
create  table if not exists nyctaxi_downsampled_dataset_testNEW (
medallion string,
hack_license string,
vendor_id string,
rate_code string,
store_and_fwd_flag string,
pickup_datetime string,
dropoff_datetime string,
pickup_hour string,
pickup_week string,
weekday string,
passenger_count int,
trip_time_in_secs double,
trip_distance double,
pickup_longitude double,
pickup_latitude double,
dropoff_longitude double,
dropoff_latitude double,
direct_distance double,
payment_type string,
fare_amount double,
surcharge double,
mta_tax double,
tip_amount double,
tolls_amount double,
total_amount double,
tipped string,
tip_class string
row format delimited fields terminated by ','
lines terminated by '\\n'
stored as textfile;

Now insert contents of the join into the preceding internal table.

queryString = """
insert overwrite table nyctaxi_downsampled_dataset_testNEW
hour(t.pickup_datetime) as pickup_hour,
weekofyear(t.pickup_datetime) as pickup_week,
from_unixtime(unix_timestamp(t.pickup_datetime, 'yyyy-MM-dd HH:mm:ss'),'u') as weekday,
if(tip_amount>0,1,0) as tipped,
if(tip_amount>0 and tip_amount<=5,1,
if(tip_amount>5 and tip_amount<=10,2,
if(tip_amount>10 and tip_amount<=20,3,4)))) as tip_class
+cos(pickup_latitude*radians(180)/180)*cos(dropoff_latitude*radians(180)/180)*pow(sin((dropoff_longitude-pickup_longitude)*radians(180)/180/2),2))) as direct_distance,
rand() as sample_key

from trip
where pickup_latitude between 30 and 90
    and pickup_longitude between -90 and -30
    and dropoff_latitude between 30 and 90
    and dropoff_longitude between -90 and -30
from fare
on t.medallion=f.medallion and t.hack_license=f.hack_license and t.pickup_datetime=f.pickup_datetime
where t.sample_key<=0.01

After a while, you can see that the data has been loaded in Hadoop clusters:

queryString = """
    select * from nyctaxi_downsampled_dataset limit 10;
pd.read_sql(queryString, connection)

Top rows of data from the table

Read data from HDI by using Azure Machine Learning Studio (classic): reader module

You can also use the reader module in Azure Machine Learning Studio (classic) to access the database in a Hadoop cluster. Plug in the credentials of your HDI clusters and Azure storage account to enable building machine learning models by using a database in HDI clusters.

Reader module properties

You can then view the scored dataset:

View scored dataset

Azure SQL Data Warehouse and databases

Azure SQL Data Warehouse is an elastic data warehouse as a service with an enterprise-class SQL Server experience.

You can provision your Azure SQL data warehouse by following the instructions in this article. After you provision your SQL data warehouse, you can use this walkthrough to do data upload, exploration, and modeling by using data within the SQL data warehouse.

Azure Cosmos DB

Azure Cosmos DB is a NoSQL database in the cloud. You can use it to work with documents like JSON, and to store and query the documents.

Use the following prerequisite steps to access Azure Cosmos DB from the DSVM:

  1. The Azure Cosmos DB Python SDK is already installed on the DSVM. To update it, run pip install pydocumentdb --upgrade from a command prompt.

  2. Create an Azure Cosmos DB account and database from the Azure portal.

  3. Download the Azure Cosmos DB Data Migration Tool from the Microsoft Download Center and extract to a directory of your choice.

  4. Import JSON data (volcano data) stored in a public blob into Azure Cosmos DB with the following command parameters to the migration tool. (Use dtui.exe from the directory where you installed the Azure Cosmos DB Data Migration Tool.) Enter the source and target location with these parameters:

    /s:JsonFile /s.Files:https://cahandson.blob.core.windows.net/samples/volcano.json /t:DocumentDBBulk /t.ConnectionString:AccountEndpoint=https://[DocDBAccountName].documents.azure.com:443/;AccountKey=[[KEY];Database=volcano /t.Collection:volcano1

After you import the data, you can go to Jupyter and open the notebook titled DocumentDBSample. It contains Python code to access Azure Cosmos DB and do some basic querying. You can learn more about Azure Cosmos DB by visiting the service's documentation page.

Use Power BI reports and dashboards

You can visualize the Volcano JSON file from the preceding Azure Cosmos DB example in Power BI Desktop to gain visual insights into the data. Detailed steps are available in the Power BI article. Here are the high-level steps:

  1. Open Power BI Desktop and select Get Data. Specify the URL as: https://cahandson.blob.core.windows.net/samples/volcano.json.
  2. You should see the JSON records imported as a list. Convert the list to a table so Power BI can work with it.
  3. Expand the columns by selecting the expand (arrow) icon.
  4. Notice that the location is a Record field. Expand the record and select only the coordinates. Coordinate is a list column.
  5. Add a new column to convert the list coordinate column into a comma-separated LatLong column. Concatenate the two elements in the coordinate list field by using the formula Text.From([coordinates]{1})&","&Text.From([coordinates]{0}).
  6. Convert the Elevation column to decimal and select the Close and Apply buttons.

Instead of preceding steps, you can paste the following code. It scripts out the steps used in the Advanced Editor in Power BI to write the data transformations in a query language.

    Source = Json.Document(Web.Contents("https://cahandson.blob.core.windows.net/samples/volcano.json")),
    #"Converted to Table" = Table.FromList(Source, Splitter.SplitByNothing(), null, null, ExtraValues.Error),
    #"Expanded Column1" = Table.ExpandRecordColumn(#"Converted to Table", "Column1", {"Volcano Name", "Country", "Region", "Location", "Elevation", "Type", "Status", "Last Known Eruption", "id"}, {"Volcano Name", "Country", "Region", "Location", "Elevation", "Type", "Status", "Last Known Eruption", "id"}),
    #"Expanded Location" = Table.ExpandRecordColumn(#"Expanded Column1", "Location", {"coordinates"}, {"coordinates"}),
    #"Added Custom" = Table.AddColumn(#"Expanded Location", "LatLong", each Text.From([coordinates]{1})&","&Text.From([coordinates]{0})),
    #"Changed Type" = Table.TransformColumnTypes(#"Added Custom",{{"Elevation", type number}})
    #"Changed Type"

You now have the data in your Power BI data model. Your Power BI Desktop instance should appear as follows:

Power BI Desktop

You can start building reports and visualizations by using the data model. You can follow the steps in this Power BI article to build a report.

Scale the DSVM dynamically

You can scale up and down the DSVM to meet your project's needs. If you don't need to use the VM in the evening or on weekends, you can shut down the VM from the Azure portal.


You incur compute charges if you use just the shutdown button for the operating system on the VM.

You might need to handle some large-scale analysis and need more CPU, memory, or disk capacity. If so, you can find a choice of VM sizes in terms of CPU cores, GPU-based instances for deep learning, memory capacity, and disk types (including solid-state drives) that meet your compute and budgetary needs. The full list of VMs, along with their hourly compute pricing, is available on the Azure Virtual Machines pricing page.

Similarly, your need for VM processing capacity might diminish. (For example: you moved a major workload to a Hadoop or Spark cluster.) You can then scale down the cluster from the Azure portal and go to the settings of your VM instance.

Add more tools

Tools prebuilt into the DSVM can address many common data-analytics needs. This saves you time because you don't have to install and configure your environments one by one. It also saves you money, because you pay for only resources that you use.

You can use other Azure data and analytics services profiled in this article to enhance your analytics environment. In some cases, you might need additional tools, including some proprietary partner tools. You have full administrative access on the virtual machine to install new tools that you need. You can also install additional packages in Python and R that are not pre-installed. For Python, you can use either conda or pip. For R, you can use install.packages() in the R console, or use the IDE and select Packages > Install Packages.

Deep learning

In addition to the framework-based samples, you can get a set of comprehensive walkthroughs that have been validated on the DSVM. These walkthroughs help you jump-start your development of deep-learning applications in domains like image and text/language understanding.

  • Running neural networks across different frameworks: This walkthrough shows how to migrate code from one framework to another. It also demonstrates how to compare models and runtime performance across frameworks.

  • A how-to guide to build an end-to-end solution to detect products within images: Image detection is a technique that can locate and classify objects within images. This technology has the potential to bring huge rewards in many real-life business domains. For example, retailers can use this technique to determine which product a customer has picked up from the shelf. This information in turn helps stores manage product inventory.

  • Deep learning for audio: This tutorial shows how to train a deep-learning model for audio event detection on the urban sounds dataset. It also provides an overview of how to work with audio data.

  • Classification of text documents: This walkthrough demonstrates how to build and train two neural network architectures: Hierarchical Attention Network and Long Short Term Memory (LSTM) network. These neural networks use the Keras API for deep learning to classify text documents. Keras is a front end to three of the most popular deep-learning frameworks: Microsoft Cognitive Toolkit, TensorFlow, and Theano.


This article described some of the things you can do on the Microsoft Data Science Virtual Machine. There are many more things you can do to make the DSVM an effective analytics environment.