Tensorboard Introduction | Deep Learning Tutorial 16 (Tensorflow2.0, Keras & Python)

Often it becomes necessary to see what's going on inside your neural network. Tensorboard is a tool that comes with tensorflow and it allows you to visualize neural network as well as how it trains itself. This tool is very helpful in debugging issues too. We will use the notebook that we created in previous videos for recognizing handwritten digits and visualize accuracy,loss with every epoch using tensorboard. We will also look at the visual graphs of neural network along with some internal computations.

🔖 Hashtags 🔖
#tensorboard #tensorboardtutorial #tensorboardgraph #deeplearningtutorial #tensorboardpytorch #pytorchtensorboard

[armagaan]

The problem you are facing at 9:20 is because you did not make a separate log for each run. You ran the code multiple times with the log_dir name being the same. You should append a unique number (I append time) to make the log-name of each run unique. If you don't, then data from every new run is saved on the previous log itself. And it does not overwrite your old data of that log (it's like adding a sentence to an already written paragraph).

At the end of the first run, you were at say, 90% accuracy then you re-run the code (without changing the log-name) and the accuracy of your first epoch goes back to 1%. That's why you see that zig-zag in the accuracy and loss graphs. It appended the 1% info in front of 90% in the log_dir. (not to mention that it saved the x-axis data too making the graph of 1% start from step-1 on the x-axs)

[harshsrivastava]

if you are facing error in tensorboard dashboard, try it in powershell prompt, move to same directory add that same commad, it worked for me !

[kunalrastogi]

unexpected very detailed graph is displayed on my tensorboard

[arjunvarmamaths]

I'm Bing watching ur deep learning videos ❤❤

[dec]

It was an OK video, however and for some reason, I cannot get Tensorboard to work from Git Bash, only using Jupyter Notebooks. A bit more focus there would have been nicer.

[陳翰儒-dm]

This is really useful, thank you sir.

[phungtruong]

Thank you for declaring hyperparameters and parameters. I always confused them

[pallavipriyadarshini]

Thank u sir.. I am doing project on CNN.. Using tensorflow

[pallavipriyadarshini]

Thank u so much sir.. Brilliant representation...

[rogerwilcoshirley]

for the line graphs - what are the light colored / shadow lines representing, i dont think you mentioned.

[adityashinde]

This is going to be really helpful to understand what goes on under the hood

[bapparoy]

Can you please detail a bit on mages in Tesorboard ? How to visialize the worngly predicted images in TB...

[AlonAvramson]

Thank you! BTW, did you also do a video on time-series forecasting?

[ntuppuppakkoranendarnnu]

Sir is there any way we can get information about dataset used from TensorBoard?

[harishsenapathy]

Nice tutorial. Is it possible to somehow view these charts during execution? For very long executions, it will be helpful to view the progress while the training is underway. Is it OK to use 'Sigmoid' on a Dense layer with 10 neurons? My understanding is that Softmax is more appropriate for multi-class classification with more than 2 classes.

[saibalaji]

Hello sir can you make a video on how to add meta data to tensor flow models to use it in android as tflite

[shravankalyankar]

How the histogram is will be giving us more information for debugging..

[ankitrawat]

ERROR: Could not find `tensorboard`. Please ensure that your PATH
contains an executable `tensorboard` program, or explicitly specify
the path to a TensorBoard binary by setting the `TENSORBOARD_BINARY`
environment variable.
This error coming when i run the command
%load_ext tensorboard
%tensorboard --logdir logs/fit

[jiyabyju]

Sir, you should see my graph...it was like noodles...so I give it up..😶

[RKYT]

aren't the actual values the unsmoothed (faded) graph? So you should always set smoothing to 0