Live visualization of the training of Keras models
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
kerasvis
.gitignore
Dockerfile
LICENSE
MANIFEST.in
README.md
docker-compose.yml
loss_accuracy_screenshot.png
setup.py

README.md

Keras live visualization

This package visualizes ongoing Keras optimizations live in your browser. Optimization runs are logged in a database. To use this visualization you have to add DBLogger as callback to your optimization.

Installation

Kerasvis is developed for Python 3 only. It does not work with Python 2. There are no plans to add Python 2 support.

Install via pip the kerasvis package

pip install kerasvis

or install directly from github:

pip install git+git://github.com/neuralyzer/kerasvis.git 

Quickstart example

from kerasvis import DBLogger
from keras.models import Sequential
from keras.layers import Dense

import scipy as sp
X = sp.rand(200, 20)
y = (sp.rand(200, 1) < .5).astype(int)

model = Sequential()
model.add(Dense(200, input_dim=20, activation="sigmoid"))
model.add(Dense(1, activation="sigmoid"))
model.compile("sgd", "binary_crossentropy")

logger = DBLogger(comment="An example run")
history = model.fit(X, y, nb_epoch=100, batch_size=64, verbose=0,
                    validation_split=0.2, callbacks=[logger]

Check the docstring of the DBLogger class for more information.

Start the keras visualization server with

export FLASK_APP=kerasvis.runserver
flask run

You'll see a web output similar to the one below

Loss Accuracy screenshot

Delete runs from the database

Runs can be deleted with the DBLogger. To delete the run with id 3 you do

db_logger = DBLogger(id=3)
db_logger.delete()

you will then be asked if you really want to delete and have to confirm with "yes".

Runs can also be delted via the web interface. To disable deletion via the webinterface set the environment variable

KERASVIS_ALLOW_DELETE=False