#1 PyTorch Book Thread: Sunday, 29 Aug 8 AM PT

YT Link: PyTorch Book Reading - 1. PyTorch basics, Tensors & NNs in PyTorch - YouTube

Hey Everyone!
This thread is for discussing, Q&A and everything else for the #1 meetup of the book reading group.

The plan is to follow an independent number of chapters wrt to the number of meetings. That being said, I estimate the study group will take ~10 weeks for the first run :slight_smile:

For our first week, it’ll be slightly theory & discussion heavy, we’ll look at Ch 1-3

Link to sign up



Will we discuss the first chapter on 29th Aug or it’s for the discussing the structure of reading group? I am looking forward to the session either way. Thank you very much!

1 Like

Thanks for checking! It’ll be the first few chapters :slight_smile:

See you then!

1 Like

What chapers to read for todays discussion?

1 Like

Chapter 1-3, depends on how much Chai I consume :smiley:


Agenda has been added.
We might make it to Ch-4’s initial bits but I would rather start slow & nicely rather than rush through the bits since I anticipate a few folks totally new to DL would also join so its important to pay attention to the base concepts (What is a tensor, what is dl, etc).


How long have you all been working in PyTorch?


8 weeks partially along with fastAI and since 2 days full time. :smiley:

3 months, but not consistently, just know till transfer learning (images).- Bare Basics.

1 year after tensorflow frustrated me

1 week, beginner in pytorch

  1. Collect data and label the items it contains
  2. Select a pretrained model for beginning
  3. Finetune the model to detect the labels which you are interested in

Apply some preprocessing like filters to find edges
Flatten the image/output of preprocessing to a single dimension vector
Use the above vector using traditional ML like RandomForest or LogisticRegression to map this vector to labels

1 Like

May be around 2 years. But I haven’t used it on a daily basis.

Numpy will not work on GPU! right ?

…and clean the data!


Load data into memory → Pre-process data → Train model → “Deploy” model → $$$

  1. Preprocess the data
  2. Model Training → Evaluation → Metrics → Loop over for accuracy
  3. Post-process → create APi’s around it for users.

*Collect Labelled Data (hopefully clean)

  • Optional - Visualization
  • Create a model
  • Train the model by feeding in the data (maybe after train + val+ test split)
  • Predict!/ Deploy
  • Retrain the loop
  • Collect data for your task.
  • Clean/preprocess data.
  • Select a model and then train the model.
  • Test the model.
    – Iterate till the model performance is good.
    → Deploy and have a data loop. ( Monitoring )
  1. Get the data
  2. Clean the data
  3. Select a model
  4. Train/fine-tune model
  5. Iterate over and over again
  6. Make predictions