Deep Learning with Natural Language Processing for SLE Classification

Stacked Denoising Autoencoders

Word Vectors

cTAKES information

9011286 - no notes

Using it

Download it from the given repo on the ctakes site (ctakes-ytex branch) Follow instructions on the site for UMLS and all that fun stuff navigate to ctakes-distribution and extract the bin tarball Use that directory as your ctakes_home environment variable

Processing Lots of Notes

Open up a command prompt or a mysql command line client and punch in “set global max_allowed_packet=102410241024;” and then leave that command line up when using ctakes

Exporting Data (CUIs) to Sparse Matrix (or ARFF)

Create a file like this:

sle-cui -1 and d.analysis_batch = '102014-3' group by f.noteid, code ]]>

Run by typing in:

cd %CTAKES_HOME% bin\setenv.bat java -cp %CLASSPATH% -Dlog4j.configuration=file:%CTAKES_HOME%\config\log4j.xml -Xmx2g org.apache.ctakes.ytex.kernel.SparseDataExporterImpl -prop %CTAKES_HOME%\data_and_exports\new_data_allnotes.xml -type sparsematrix

sparsematrix can be replaced with “weka” to get a .arff file

Sparsematrix Type: Produces 3 files: attributes.txt List of all the attributes in the sparse matrix Added subject_id to this list instance.txt A list of instanceid’s along with the patient diagnosis data.txt A list with 3 columns: Instance_id produced by ctakes (starts at 1) Sparse Matrix Entry being specified (CUI) Amount of CUI confirmed (also includes noteid’s as first value)

Theano (might be able to just use the pip easy install now)

Paste this in a file called ".theanorc.txt" ('.txt' optional) at the C:/ level

[blas]

ldflags = -LC:\Users\Clayton\AppData\Local\Enthought\Canopy\App\appdata\canopy-1.5.2.2785.win-x86_64\Scripts -lmk2_core -lmk2_intel_thread -lmk2_rt

[gcc]

cxxflags = -IC:\MinGW\include