Fairly large Pull Request adding a new jupyter notebook that trains two distinct two hidden layer Tybalt models. The notebook is added in e82c365 and nbconverted script in 61bbb1c. The biggest change is the addition of a Tybalt class that does most of the heavy lifting. This partially addresses #74 but not completely since the class cannot be imported yet.
The two models have architectures:
5000 -> 100 -> 100 -> 100 -> 500
5000 -> 300 -> 100 -> 300 -> 500
They are trained in e82c365 with the optimal hyperparameters uncovered in #71 and #77
The rest of the files are all output files generated from the notebook including:
Description
Format
Git LFS?
Commit
The actual encoder and decoder keras models
.hdf5
:white_check_mark:
9883611
The latent feature matrices of both models
.tsv.gz
:white_check_mark:
3dce40f
The gene weight matrices for both models
.tsv
:white_check_mark:
3cc4e9b
Training history figure + architecture
.png/.pdf
:x:
f346f65
I also update the git LFS .gitattributes file in 46f113f to automatically track any .hdf5 file saved in the models/ folder (shout out to @dhimmel here! :smiling_imp: ).
Fairly large Pull Request adding a new jupyter notebook that trains two distinct two hidden layer Tybalt models. The notebook is added in e82c365 and nbconverted script in 61bbb1c. The biggest change is the addition of a
Tybaltclass that does most of the heavy lifting. This partially addresses #74 but not completely since the class cannot be imported yet.The two models have architectures:
They are trained in e82c365 with the optimal hyperparameters uncovered in #71 and #77
The rest of the files are all output files generated from the notebook including:
.hdf5.tsv.gz.tsv.png/.pdfI also update the git LFS
.gitattributesfile in 46f113f to automatically track any.hdf5file saved in themodels/folder (shout out to @dhimmel here! :smiling_imp: ).