Class final project for MIT 6.S052 Spring 2024 on classifying EMG signals
2048 from Muedit (calculated 2000 from S1_10_DF.otb%2B_decomp.mat_edited.mat)
4 grids of 64 electrode for KE (256) 265 for DF
- cnn.py is more like a library
- run_model to run the model, this is most likely going to be custom to your run so you can keep a local copy that is not pushed up to git
- exploration.ipynb, like run_model you can keep a local version of this, this what I used to understand the dataset
look at run model file, Weights in model weights, pick the one you would like, the best that I have found is already selected
test_files.txt -> these are the only file you would need to run test. You can take them from np_emp_dataset
make sure you are only loading these files, by commenting out the dataset and dataloader for training
these corrospond to losses while training the model. Each file contains a list. there is one entry for every epoch in the list. The timestamp in the names of the files matches the training run and corrosponding set of modelweights.
model was trained until 1000 epoch, then a second run starting at 1000 epoch upto 10k was run. the losses that corrospond to this training history are split into two files. 03-05-24-10_54__losses.txt 03-05-24-13_23__losses.txt