Track real-time metrics of TensorFlow Model during training using Notifly

Notifly is a Pypi package designed to track the model metrics during real-time training using a wrapper over Tensorflow callbacks, which plots the accuracy and loss over each epoch. Notifly also tracks the system resources over the runtime thus providing details about the training in runtime, It can also share the details over Discord, Telegram, Teams and Slack.

Built with Python 3

Prerequisites:

• Python It is preinstalled in Ubuntu 20.04. To check the version use command:

  python3 --version
  

  If it is not preinstalled for some reason, proceed here and download as per requirement. Run the following command in terminal to download the required packages for running the tool locally :
• Using requirements file :

pip3 install -r requirements.txt

• Directly download packages:

  pip3 install requests==2.24.0
  pip3 install matplotlib==3.2.2
  pip3 install slackclient==2.9.3
  

Install the package

Run the following terminal commands to install the package on the given distros.

• Terminal: shell

pkg install python3

pip3 install notifly

• Ubuntu/Debian

sudo apt install python3-pip

pip3 install notifly

• Arch

sudo pacman -S python3-pip

pip3 install notifly

This may take a while depending on the network speed.

Working of the tool

Telegram

To see how the tool works,

1. Create the telegram bot.

2. Getting the bot API token

   1. Search and go to _@Botfather_ .
   2. Message /mybots .
   3. Select the bot.
   4. Select the API token displayed in message.
   5. Copy and use in sample code.

   from notifly import telegram              #import the package    
   x = telegram.Notifier('bot API token')   #create object of class Notifier
   x.send_message('message')                #send message
   x.send_image("image address")            #send image(.jpg or .png format)
   x.send_file("file address")              #send document
   x.session_dump()                         #creates folder named 'downloads' in local folder, downloads/saves message,chat details for current session in 'sessio_dump.json' file
   

3. Run sample code.

Discord

To see how the tool works,

1. Create server.

2. Create and copy server webhooks instruction and use in sample code.

   from notifly import discord
   x = discord.Notifier(r'webhook')         #create object of class Notifier
   x.send_message('message')                #send message
   x.send_file("file address")              #send file
   x.send_file("image address")             #send image
   

3. Run sample code.

Slack

To see how the tool works,

1. Create app. Follow these steps,
   1. Go here to create a new API for slack.
   2. Choose to Create an App .
   3. Enter App Name and select workspace. Click Create App.
   4. Under Add features and functionality select Incoming Webhooks and Activate Incoming Webhooks.
   5. Scroll down, select Add New Webhook to Workspace and select a channel from the drop down.This channel name is used as an argument in the sample code. Click Allow.
   6. Select OAuth & Permissions from left-sidebar.
   7. Under Scopes > Bot Token Scopes click Add an OAuth Scope and add the following scopes,
      chat:write   chat:write.public   files:write   users:write
   8. Scroll up, under OAuth Tokens for Your Team copy the Bot User OAuth Access Token to use in sample code.
   9. Click Reinstall to Workspace, select channel and click Allow.

2. Write sample code.

   from notifly import slack
   x= slack.Notifier('token', channel='channel-name')      #create object of class Notiflier
   x.send_message('message')      #send message
   x.send_file("image or file address")      #send image/file
   

3. Run sample code.

Tensorflow Integration

Plug and play feature for your tensorflow callbacks

# create your notifier using above methods
from notifly import discord
notifier = discord.Notifier(r'webhook') 
class MyNotifierCallback:

    @notifier.notify_on_epoch_begin(epoch_interval=1, graph_interval=1, hardware_stats_interval=1)
    def on_epoch_begin(self, epoch, logs=None):
        pass

    @notifier.notify_on_epoch_end(epoch_interval=1, graph_interval=1, hardware_stats_interval=1)
    def on_epoch_end(self, epoch, logs=None):
        pass

    @notifier.notify_on_train_begin()
    def on_train_begin(self, logs=None):
        pass

    @notifier.notify_on_train_end()
    def on_train_end(self, logs=None):
        pass

model.fit(callbacks=[MyNotifierCallback()])

Learn more about Notifly ✨

Read the wiki pages which has all the above steps in great detail with some examples as well 🤩🎉.

Contributing

1. Fork the Project
2. Create your Feature Branch

   git checkout -b feature/mybranch

3. Commit your Changes

   git commit -m 'Add something'

4. Push to the Branch

   git push origin feature/mybranch

5. Open a Pull Request
   
   Follow the given commands or use the amazing GitHub GUI
   Happy Contributing