2.6 KiB
Setup
Backend configuration
This demo assumes that you have access to a SQL Server instance, and have already setup SQL Server R Services and R Server Operationalisation. For more information about these, see the following MSDN pages:
You'll also need to ensure that the R Server Operationalisation accounts can access the database. See your database administrator for help if you are using a shared SQL Server instance.
R configuration
The MicrosoftML package can use GPU acceleration to fit neural network models, and this is highly recommended. To enable this, see the help for MicrosoftML::rxNeuralNet. In a nutshell, you install the NVidia CUDA Toolkit 6.5 and cuDNN v2 Library, and then copy some .dlls to the MicrosoftML mxLibs directory. Currently only CUDA acceleration is supported.
The demo uses the following R packages, other than those that come with R Server:
All of these are available on CRAN, except for RMLtools which is on Github. You can install this package using devtools:
install.packages("devtools")
devtools::install_github("andrie/RMLtools")
In addition the imager package and its dependencies must also be installed into a location accessible by SQL Server R Services, so that they are available to R code running inside a stored procedure.
Project settings
The script code/settings.R is used to set project options. It will read two additional files which you should put into the code directory:
deployDbConnStr.txtcontaining the ODBC connection string for your SQL Server instance.deployCred.txtcontaining your RServe login credentials (for R Server operationalisation).
You should also edit settings.R to set the paths where you want the raw and processed galaxy images saved. The image files total about 2.4GB after processing, so make sure you point this to a location that has enough space.
Finally, run the script code/dataprep/initDataDirs.R. This will create the data directory structure in which model objects, R data frames etc are stored.