Name: hylas
Owner: Data Science for Social Good
Description: Webapp for visualizing ML'd data
Created: 2015-10-21 17:46:21.0
Updated: 2016-01-12 20:15:48.0
Pushed: 2016-01-15 18:35:59.0
Homepage: null
Size: 0
Language: JavaScript
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
Hylas is a dashboard for performing and visualizing the results of machine learning experiments. With Hylas, you can:
The first step in analyzing data is to upload a CSV file that you want analyzed. The CSV file should:
You can specify which columns signify the unit id and the label by filling out the “Unit ID column” and “label column” text boxes, respectively.
You can also select the classifiers and hyperparameters to use at the bottom of the page. Use the “+” buttons to add a classifier, parameter, or parameter value, and the “-” buttons to remove them.
Once you have selected your CSV and your classifiers, you can press the “Submit” button.
In addition to uploading a CSV, you can also upload a pickled Diogenes Experiment object (see Exporting below)
After you have uploaded some data, you can move on to the “Report” page. This should happen automatically. You can also click on “Report” in the upper right.
First, under the “Models” heading, click on the classifier that you are interested in studying. That will bring you to the “Model Performance” heading.
The “Model Performance” tab will show you general metrics about how well the classifier made predictions about your data. The “Top Units” tab tells you which units were most likely to be labeled as positive and what their most predictive features were. Click on one of these units to get more detail.
On the top of the “Unit Performance” section are the values for the given unit for the most important features. When you click on a feature, the “Distributions” tab shows the distribution of this feature for positive and negative results in the data set, as well as the value for that feature for the unit that we selected. The “Similar” tab shows which units are most similar to the selected unit.
Click on “Export” in the upper right to see different ways to export the data. The PDF and CSVs give detailed summaries of how well the different classifiers performed. They correspond to the PDF and CSV reports generated by the Diogenes Experiment object. The pickled Experiment is the Experiment object itself, which you can either unpickle and examine yourself (requires Diogenes) or you can Upload into Hylas in case you would like to look at the analysis again.
bower install
.config.py
file in the project's root directory. You may use the
included config.py.sample
as a template.python server.py
to start the serverhttp://127.0.0.1:5000/
with a web browser. This creates the user
database.python add_user.py USERNAME PASSWORD
to add a user to the database.python server.py
.http://127.0.0.1:5000/
and log in with the username and password you have just created.