morvanzhou/pytorch-tutorial

★ 8,455Jupyter NotebookAudience · researcherComplexity · 2/5Setup · moderate

Mindmap

mindmap
  root((pytorch-tutorial))
    Fundamentals
      NumPy relationship
      Variables and gradients
      Activation functions
    First networks
      Regression
      Classification
      Optimization strategies
    Advanced architectures
      CNN image tasks
      RNN sequences
      GAN generative
      Autoencoder
    Practical techniques
      Dropout
      Batch normalization
      GPU training

mindmap root((pytorch-tutorial)) Fundamentals NumPy relationship Variables and gradients Activation functions First networks Regression Classification Optimization strategies Advanced architectures CNN image tasks RNN sequences GAN generative Autoencoder Practical techniques Dropout Batch normalization GPU training

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Learn how neural networks work by running standalone PyTorch scripts covering regression, classification, and model saving.

USE CASE 2

Build a convolutional or recurrent neural network by following along with the provided Jupyter notebook examples.

USE CASE 3

Use the GAN or autoencoder example code as a starting template for your own generative model project.

Tech stack

PythonPyTorchJupyter NotebookNumPy

Getting it running

Difficulty · moderate Time to first run · 30min

Requires Python with PyTorch and Jupyter installed, GPU is optional but needed for the GPU-specific tutorial sections.

In plain English

This repository is a set of worked code examples and Jupyter notebooks for learning PyTorch, a popular Python library for building neural networks. The tutorials start from the basics and progress through increasingly advanced network designs. A companion Chinese-language video and text course is available at the author's website for Mandarin speakers. The content is organized into four sections. The first covers PyTorch fundamentals: how it relates to NumPy (a numerical computing library), how variables and gradients work, and how to apply activation functions. The second section walks through building your first networks, covering regression (predicting a number), classification (sorting inputs into categories), saving and loading trained models, training in mini-batches, and comparing different optimization strategies. The third section introduces more advanced architectures: convolutional neural networks (CNNs, commonly used for image tasks), recurrent neural networks (RNNs, useful for sequences like text or time series), autoencoders (networks that learn compressed representations of data), generative adversarial networks (GANs, which generate new data that resembles a training set), and reinforcement learning approaches. The fourth section covers practical techniques like dropout (a method to reduce overfitting), batch normalization, and running training on a GPU. Each topic has a standalone Python script you can run directly, and many also have Jupyter notebook versions contributed by a community collaborator. The tutorials assume some familiarity with Python but explain each neural network concept as it appears in the code.

Copy-paste prompts

Prompt 1

I want to follow the pytorch-tutorial regression example. Explain what the training loop does step by step and what I should see printed when I run it.

Prompt 2

Using the RNN example in morvanzhou/pytorch-tutorial, help me adapt it to predict the next word in a sentence using my own text dataset.

Prompt 3

I want to run the CNN image-classification notebook from this tutorial. What dependencies do I need to install and how do I launch Jupyter?

Prompt 4

The tutorial covers GANs. Walk me through what the generator and discriminator networks in the example are doing and how I would swap in my own image data.

Prompt 5

Help me adapt the batch normalization example from this tutorial to add dropout as well, following the style of code used in the repo.

Open on GitHub → Explain another repo

← morvanzhou on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.