From 32ca971fec1dca345088a5e4fd7685e3bc7e1926 Mon Sep 17 00:00:00 2001 From: Dawith Lim Date: Tue, 21 Oct 2025 13:25:41 -0400 Subject: [PATCH] Update README.md --- README.md | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/README.md b/README.md index f3baf15..ae85bd7 100755 --- a/README.md +++ b/README.md @@ -15,6 +15,10 @@ To enter the docker container, run `launch.sh` for Linux or `winlaunch.ps1` for Windows. Once inside, change directory to `/app/code` and run the desired script. +### Data inspection +Running `python inspect_data.py` produces plots that offer insights about the distribution of values for the data. This should be used to make informed judgment on the regularization parameters for each data. +The reason this is not done adaptively is because the `timeseries` blocks combine multiple data with different distributions, and each of them can drift significantly experiment to experiment. In other words the data is likely not nicely unimodal, and some preprocessing may be needed to filter out extreme outliers. The data inspection tool is there for the first step: See the data distribution, and spot the outliers. + ### Encoding Required Inputs: .mat files in the matfiles directory, .hdf5 file in hdf5 directory, or .csv files in csv directory. @@ -26,6 +30,12 @@ Model inputs (x): Encoder output (y): * Two 1xN one-hot encoded classification vectors that predicts sample type and treatment applied. +To run encoding, use `python train.py`. This is subject to change later when the decoder is actually operational and there is a need to distinguish the training code file names. + ### Decoding Work in progress + +### Model evaluation tool + +Run `python model_eval.py` to run the model metric evaluation script. It pulls the hyperparameters and test set prediction metrics from `metrics.csv` and plots the two metrics (accuracy and loss) against all pairwise permutations of hyperparameters. In every plot, one hyperparameter is the x-axis and the other hyperparameter is the hue.