PerceptiLabs
Search…
Changelog
This page provides a changelog that lists new releases and updates to the PerceptiLabs products and services.

Free Version

Note
To upgrade to the latest version of PerceptiLabs, run:
$ pip install --upgrade perceptilabs
To install or upgrade to the latest nightly build of PerceptiLabs, run:
$ pip install --upgrade pl-nightly

Version 0.11.15 (April 20, 2021)

Bugfixes
    A few components (such as DataRandom) which were previously not working have now been fixed.

Version 0.11.13 (April 13, 2021)

Features
    PerceptiLabs is now running TensorFlow 2.4! When loading models from previous versions of PerceptiLabs, be sure to review your model’s components, as some may not be compatible with this newer version.
Bugfixes
    The frontend will no longer slow down just because you load a (reasonably) large dataset or add more layers.

Version 0.11.12

This version number was skipped. See Version 0.11.13 above.

Version 0.11.11 (Mar 24, 2021)

Features
    The custom code editor has been updated. It's now based on Visual Studio Code which is much easier to use and supports things such as searching, replacing, highlighting keywords etc.
    You can now also minimize and maximize the code window.
    The charts have been improved to be interactive. You can now hover the mouse over the charts to see specific values, as well as zoom in and out on the chart using the scroll wheel.
    You can now place multiple components after each other by holding the Shift button. If you place one close enough to an already existing component, it will automatically connect to that component.
    There is now a feature map slider on the workspace. By hovering over a convolutional component, you can go through all the feature maps of that component.
    The Grid functionality has been added; you can now enable a grid through the bottom left checkbox.
    Create Model has been given a face lift; you now get a preview of the templates before you choose to create them.
    The tool now supports 'offline' mode, provided that you have logged in at least once before.
Bugfixes
    The Loss and Accuracy now show average values rather than the latest value, giving a much better estimate of how well the training is going.
    The workspace grid has now been improved so that components will be placed in line with each other, regardless of whether you are zoomed in or out.
    Copy, paste, and cut functionality has been improved. Whenever performing these actions, the connections will now be pasted correctly.
    There has also been a lot of performance improvements, making sure that the tool runs smoother.

Version 0.11.10 (Mar 2, 2021)

Bug Fixes
Fixes a fatal error coming from a dependency.

Version 0.11.9 (Feb 23, 2021)

Bugfixes
Hiding some Kernel messages which used to overflow the terminal.

Version 0.11.8 (Feb 15, 2021)

Bugfixes
Exported models now retain their weights, which previously did not follow in the export.

Version 0.11.7 (Jan 21, 2021)

Bugfixes
Save and Save As now work as intended.

Version 0.11.6 (Jan 6, 2021)

Bugfixes
Fixed two fatal errors.

Version 0.11.5 (Nov 20, 2020)

New Functionality
    Training layers now show a preview.
    There is now a warning popup when you attempt to delete models.
    The optimizer Adagrad has been added to the Classification Training component.
Bugfixes
    Autosettings notification is not shown for data component.
    Now installs the correct NumPy version.
    The component settings are automatically scrolled to the top now when changing components.

Version 0.11.4 (Nov 11, 2020)

New Functionality

    The Model Editor now features a map of the model in the bottom left corner.
    Users can now scroll the model Using Ctrl+drag.
    Component settings are locked after editing code.

Version 0.11 Silver (Oct 6, 2020)

New Functionality

    We have made a number of significant UI improvements to streamline the workflow
    The new "Model autoconfig" feature automatically chooses "good" hyperparameters as you build your model
    Models can now be shared to GitHub through PerceptiLabs
    Both the front end (UI) and Kernel are now part of the same Python package that is installed and run on your local machine
    Components have been added/modified:
      Local(formerly Data): the Load Dataparameter can only be used to load files or select folders. Environments are now selected using the new Environmentcomponent
      Switch: the Selected layerparameter has been removed
      Classification(formerly Normal): the Labels parameter has been removed and Additional Stop Conditionadded
      Regression: the Labelsparameter has been removed
      GAN: Additional Stop Condition parameter has been added
      Object Detection: the Labelsparameter has been removed and Additional Stop Conditionadded
    We have switched to a new user authentication system, so existing users will need to create a new user account in order to use PerceptiLabs.

Version 0.10 (Jul 14, 2020)

New Functionality

    Both the front end (UI) and Kernel are now part of the same Python package that is installed and run on your local machine
    There is now a new Models view screen as well as a grouping of your models into Projects! This will give you a clear overview of which models you have, which are currently training as well as how the last training attempt went for your model. This also comes with an update to the design of the current views.
    You can now export as a Jupyter Notebook file.
    Kernel v2 is now shipped:
      The main purpose of v2 is to have a much more flexible and robust framework to work with. To start with this is going to be apparent as a much more structured way to work with custom code. But will later on open up for more features as well.
      Safer decoupling between the training layer and the Kernel. This means that as long as you use our interfaces, it does not matter what model of framework you are building, visualizations will still work.
      Changes how the code is constructed inside the components and allows for much more flexible models because the training layer having full control over the graph.
    GAN has been added
    Object Detection has been added

Version 0.9 (Feb 27, 2020)

Moved to web and optimization

    We have now moved over to a web based tool! To use it, you just need to install the PerceptiLabs pip package and then go to https://ml.perceptilabs.com/. This means that you will run our Kernel in your local environment so you have full access to any dependency you wish to install. Go here for more details: https://perceptilabs.com/docs/installation
    You can now read CSV files which are larger than your memory allows.
    The preview will now display multiple times faster when loading heavy components (such as big data components or components with many parameters)
    New file chooser design, now works much better for the browser

Version 0.1.8 (Dec 17, 2019)

Fixing bugs and optimization

    The program will now visualize previews in layers much faster
    The program will now always turn off the core properly
    Preview will now always show some variable, even if 'Y' isn't defined
    Export and Save Trained network is now working
    LayerContainers can now contain LayerContainers
    Data Environment layer now properly shows Action Space
    Training Layer for RL now have the additional options Episodes and Max Steps

Version 0.1.7 (Dec 3, 2019)

Fixing bugs

When grouping several components, i.e. creating a layer container, we fixed an issue where you sometimes were not able to copy or delete specific layers, as well as making sure that copying an entire layer container now works as intended.
We have made sure that you now can delete several components on the workspace at the same time. In the Tutorial section, some users got the issue of the training not properly starting. which now is fixed.
We have improved the copy paste functionality of layers so that they now will be pasted in proximity to the mouse cursor, rather than the original layers positions.

Version 0.1.4 (Nov 27, 2019)

Read data larger than RAM

We have removed the bug which made it seem like you never got passed the splash screen.
Now you can read and use larger datasets that doesn't fit into RAM.
There is a small design change in the Data component - if you use a CSV file.

Version 0.1.3 (Nov 25, 2019)

Fixing bugs

In this release we have fixed some bugs. For those of you who have used earlier versions than v0-1-2 - If you get a black screen when trying to start PerceptiLabs, you need to delete the following folder:
C:\Users\USER\AppData\Roaming\PerceptiLabs
or
/Users/USER/Library/Application Support/PerceptiLabs
Last modified 5mo ago