Using the Geo Artificial Intelligence Data Science Virtual Machine
Use the Geo AI Data Science VM to fetch data for analysis, perform data wrangling, and build models for AI applications that consume geospatial information. After you've provisioned your Geo AI Data Science VM and signed in to ArcGIS Pro through your ArcGIS account, you can start interacting with ArcGIS desktop and ArcGIs online. You can also access ArcGIS from Python interfaces and an R language bridge that's preconfigured on the Geo-Data Science VM. To build rich AI applications, combine the Geo-Data Science VM with the machine-learning and deep-learning frameworks and other data science software that are available on it.
The Python library, arcpy, which is used to interface with ArcGIS, is installed in the global root conda environment of the Data Science VM that's found at
- If you're running Python at a command prompt, run
activateto activate into the conda root Python environment.
- If you're using an IDE or Jupyter notebook, you can select the environment or kernel to make sure you're in the correct conda environment.
The R bridge to ArcGIS is installed as an R library named arcgisbinding in the main Microsoft Machine Learning Server standalone instance that's located at
C:\Program Files\Microsoft\ML Server\R_SERVER. Visual Studio, RStudio, and Jupyter are already preconfigured to use this R environment and will have access to the
arcgisbinding R library.
Geo AI Data Science VM samples
In addition to the machine-learning and deep-learning framework-based samples from the base Data Science VM, a set of geospatial samples is also provided as part of the Geo AI Data Science VM. These samples can help you jump-start your development of AI applications by using geospatial data and the ArcGIS software:
Getting started with geospatial analytics with Python: An introductory sample showing how to work with geospatial data through the Python interface to ArcGIS is provided by the arcpy library. It also shows how to combine traditional machine learning with geospatial data and then visualize the result on a map in ArcGIS.
Pixel-level land use classification: A tutorial that illustrates how to create a deep neural network model that accepts an aerial image as input and returns a land-cover label. Examples of land-cover labels are forested and water. The model returns such a label for every pixel in the image.
Additional samples that use the Data Science VM are available here: