cleanpegasus / tensordash

TensorDash is an application that lets you remotely monitor your deep learning model's metrics and notifies you when your model training is completed or crashed.

TensorDash

TensorDash is an application that lets you remotely monitor your deep learning model's metrics and notifies you when your model training is completed or crashed.

Go to docs

Why Tensordash?

1. Watch your model train in real-time.
2. Supports all major deep learning frameworks.
3. Remotely get details on the training and validation metrics.
4. Get notified when your model has completed trainng or when it has crashed.
5. Get detailed graphs on your model’s metrics.

Installation

Installing the Python Package

There are two ways to install tensordash:

• Install tensordash from PyPI (recommended):

Note: These installation steps assume that you are on a Linux or Mac environment. If you are on Windows, you will need to remove sudo to run the commands below.

sudo pip install tensor-dash

If you are using a virtualenv, you may want to avoid using sudo:

pip install tensor-dash

• Alternatively: install tensordash from the GitHub source:

First, clone TensorDash using git:

git clone https://github.com/CleanPegasus/TensorDash.git

Then, cd to the TensorDash folder and run the install command:

cd TensorDash
sudo python setup.py install

Installing the Android App

Install the android app from the play store.

How to use

Tensorflow

Keras

PyTorch

Fast.ai

We are currently working on:

• Development for iOS app
• Addition of custom metrics to the model
• Auto-update the dashboard screen on Android App

Have feedback or a feature request? Drop us a mail at [email protected] or raise an issue on Github

tensordash's Issues

Accuracy Not Displayed in App

Saw this in Reddit and thought it was cool, so wanted to give it a try, The Loss function was correct but the accuracy wasn't displayed.

Colab notebook with which I tested this :

https://colab.research.google.com/drive/1o_-d_vbRIpNBq1F9ljj88GLnNvPvROO-

Attaching a Picture of the App

TensorDash is no Longer on PlayStore

Its been month since the app disappeared from play store, google getting strict nowadays regarding apps that might be issue for the app takendown.

Pls Provide with apk in the release section .

thank you

Add support for multiple metrics for validation

Is your feature request related to a problem? Please describe.
My model uses multiple metrics for validation, and it would be useful if I could visualize each one of them.

Describe the solution you'd like
When sending info to server, it would be nice if a dict of metrics (composed by its names and current value) could be provided.

Describe alternatives you've considered
Maybe it would be nice to split the sendLoss function into different functions, like sendLoss, sendMetrics.

Additional context
For cases when working with Super Resolution, one should compare images using regular metrics (e.g. PSNR) and metrics that consider human perception (e.g. SSIM), for example.

Refreshing icon does not stop

If there are no projects present and you try refreshing, even the projects are refreshed, the refreshing symbol does not stop.

Add support for PyTorch Lightning

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

extending torchdash

I am looking to extend torchdash to include logging some custom variables. For example, I am trying to include coarse loss and fine loss (both in training and validation). What is the best way to accomplish this?

I am currently looking to add a new file alongside *dash.py. Do you suggest its better to extend within the file torchdash?

Suggestion: split project

I believe it would be better if the project is splitted in two: one for the Android app, another for the API. This way the bugs, feature requests, etc, would be per project part. Also each project could have its own CI, code styling, etc.

Making iOS app for TensorDash.

Create a TODO List

Hey @CleanPegasus, This project seems really cool, would like to contribute.

I created a PR ( #9 ) but that was something I felt could be done so as to not expose my password.

If you guys have a TODO list adding that to the README would help people contribute to the Repo.

Some things I can currently think of are :

• Multiple-Model Training using Single Sign-In
• Status shows running when Training is forcefully terminated before completion.
• Allow Models with the Same Name and Include Timestamps to differentiate them.
• Support for PyTorch using Fast.ai callbacks
• Support for Wear OS and Android Auto

Add support for multiple losses during training

Is your feature request related to a problem? Please describe.
My model uses multiple losses during training, and it would be useful if I could visualize each one of them.

Describe the solution you'd like
When sending info to server, it would be nice if a dict of losses (composed by its names and current value) could be provided.

Describe alternatives you've considered
Maybe it would be nice to split the sendLoss function into different functions, like sendLoss, sendMetrics.

Additional context
For cases when working with GANs, that we have two different models with different losses, or maybe even one single model with two weighted losses.

Delete Button to remove a model

Feature Request
There is no way to remove a model.

Solution
A button in every model's page to delete that particular model and redirect to the model list

Validation loss equal to training loss

Describe the bug

Torchdash is plotting same values for training and validation data.

Display training and validation loss in the same pane

It would be great to be able to see the training and validation loss on the same set of axes, so that they can be more easily compared without having to go back and forth between the two plots. This is especially true when the two y-axes are on different scales. It makes it very difficult to determine how close they are to one another.

Multiple entries of the same model

Description the bug
Multiple entries of the same model are shown on starting the app from the notification. The entries all disappear and become a single entry on refresh.

To Reproduce
Steps to reproduce the behavior:

1. Start model training
2. Tap on the notification that appears in the notification panel
3. See error

Expected behavior
A single entry for a particular model on the start up of the app initiated from the notification.

Screenshots

Smartphone (please complete the following information):

• Device: Samsung A6
• OS: Android 9
• Version 1.0

Package does not send accuracy metrics in keras 2.2.5

Google colab uses Keras 2.2.5 and tensordash package does not send accuracy metrics to firebase in this version.

ValueError: bad marshal data (unknown type code)

Describe the bug
the command
!python setup.py install is not running in colab or kaggle notebook and showing some value-error

To Reproduce
Steps to reproduce the behavior:

1. !python setup.py install
2. See error : ValueError: bad marshal data (unknown type code)

Package

• Package Version: [eg, v1.6]
• Deep learning framework used: [eg. keras, tensorflow]

Desktop (please complete the following information):

• Browser chrome