Allison Mann & John C. Merfeld's Final project for Machine Learning at Boston University
-
Make sure Python 3, dlib, imutils, opencv, and Tensorflow/Keras are installed
-
Run the pipeline script, which should handle all steps from reading the raw data to creating the model
bash pipeline.sh 100 wordFiles/approvedWords.json vidData/vidData.json modelData/x.npy modelData/y.npy 30 50 predictions/pred.csv
This will create two CSV files in the predictions folder: pred.csv and pred_ground_truth.csv. The first has the model's probability output for each class. The second has the actual class vectors.
- To evaluate the model for both topline accuracy and confusion matrix, follow the procedures in
R/confusionMatrix.R
To send changes to the cluster and test:
## local
scp src/* [email protected]:/projectnb/cs542sp/jcmerf/src/
# or
scp src/* [email protected]:/projectnb/cs542sp/jcmerf/src/
## on cluster
# load centOS 7 so we can run dlib
scc-centos7
# run scripts, e.g.
bash pipeline.sh 100 approvedWords1.json vidData1.json x1.npy y1.npy 30 50 pred1.csv
## to submit it as a batch job:
qsub pipelineBatch.sh 200 wordFiles/approvedWords1.json vidData/vidData1.json modelData/x1.npy modelData/y1.npy 30 50 predictions/pred2.csv
# you can check on progress with
qstat -u ainezm
# or
qstat -u jcmerf
# output will be printed to output.txt (I added a progress bar in some of the scripts)
# errors will be printed to error.txt
## local
scp [email protected]:/projectnb/cs542sp/jcmerf/predictions/* .
approvedWords1.json should look like this:
{
"I": 1,
"ME": 2,
"YOU": 0
}
All the other file names can just be names, they don't need to exist yet.
