SOM Graphical User Interface (GUI)

In this GUI we are going to embed the SOM workflow inside GUI to make it easier to use for people who want to use but have no experience in coding.

Fields in Data Preprocessing --> coming

1. Data Cleaning
2. Feature Extraction

Fields in SOM:

1. Training

   • Choose the mapsize and SOMFactory build hyperparameters, including:

   SOMFactory	sm.train
   data	n_job = 1
   mapsize=None	shared_memory = False
   normalizer=None	verbose = "info"
   initialization='pca'	train_rough_len
   lattice='rect'	train_rough_radiusin
   neighborhood	train_rough_radiusfin
   mask=None	train_finetune_len
   mapshape='planar'	train_finetune_radiusin
   training='batch'	train_finetune_radiusfin
   name='sompy'	train_len_factor = 1
   component_names=None	maxtrainlen = inf

2. Visualization

   • Heatmaps
   • Cluster Map
   • UMatrix Map

MVP

This MVP should include the following functions:

1. choose files from local directory (csv format only)
2. choose hyperparameters, following the table above

This MVP is built on the assumption that the csv feed is nxm matrix with valid numerical values for properties users want to train. So the entire dataset is the design matrix that does not need further processing.

Milestone

Milestone 1 (1/23/2020):

In the one_page_all.py I have designed a one-page frame for people to input hyperparameters of SOM model and click to train. I consider the output of visualizations and export them to png files. The click events need to be embedded. The tool I used is a tkinter GUI designer called PAGE, one can be found on sourceforge: http://page.sourceforge.net/.

In order to run that you need to download ActiveTcl, one can be found here: https://www.activestate.com/products/tcl/

In this repository, the user can create a working environment first, then install tkinter. One then can run command to start the GUI: python one_page_all.py

1/27/2020

For some reason jupyter does not start successfully based on the current environment.yml, but the rest of them function normally. I would ignore this issue right now and focus getting the MVP done soon.

1. I have successfully trained the model using GUI
2. I need to automate visualization generation
3. Make the progress bar work

The training phase and visualization phase are separated. After training the model we can select any of them (if you trained multiple) and generate visualization based on them.

1/30/2020

All functions work normally, I need to tune the functions for job sequence. But all functions are successfully embedded. I am not sure if tuning radius can be supported in Tim's code since I have not looked at it yet. The following is the current workflow

1. input training hyperparameters
2. click "Training" to select dataset
3. click "visualizations" to select both dataset and trained model to generate component list in the listbox and visualization maps.
4. In this case we have to generate all the way to heatmaps and close the pop up window to select component property user wants to project on the cluster map. (It is stupid, I know)
5. generate the projection map --> min-max scaling is required for numerical values in a certain range, categorical values can be used directly.

Heads Up

1. Do not install scipy, it has a conflict with existing SOMPY package.