COURSE: A deep understanding of deep learning

TEACHER: Mike X Cohen, sincxpress.com

COURSE URL: udemy.com/course/deeplearning_x/?couponCode=202212

Kaggle Dataset Link: WineQuality,MNIST

Disclaimer: I have ZERO affilation with Professor Mike nor Udemy. Just an ambitious learner ♪—— Ｏ（≧∇≦）Ｏ ——♪

Inspiration

Machine Learning is one of my career interests. In the past, I have been working on projects that involves algorithms like Linear Regressions, Decision Trees, and KNN Classifiers. Recently, I have participated in a hackathon project with my friend and got fascinated by the power of CNN in detecting plant diseases. Eventually, we won as a Best Use of AI/ML. This prompts me to learn more about deep learning.

From my perspective, Mike is a pretty cool instructor. I enjoy his straightforward explanation of the topics. I appreciate how he links the concepts to real-world coding applications. This is a great course if you are new or still building a solid foundation to deep learning!

Udemy often offers a large discount on the courses (It was 75% off when I bought it). So if my notebook sparks your interest on deep learning, feel free to check out the course!

Guide to Use this Notebook

It is recommended to use this notebook on Google Collab. There is some limitation when I ran on Jupyter Notebook (Lack free GPU and downloadable datasets).

• For Google Collabs: Don't change runtime environment to GPUs unless you see lines like device = torch.device("cuda" if torch.cuda.is_available() else "cpu"). Overusing them may lead to Google Colab restricting GPU access.

The number before each folder/file indicates its sequence in the section. I recommend you to explore those folder/file in order because many are built on top of each other.

These folders are divided into 2 categories:

1. On-paper Theory Notes: 00-paper-notes
2. Coding Notes: From 01-python to 24-RNN

Feel free to tweak the parameters to facilltate your learning experiences, especially for the milestone project folders like 14-FFNmilestone

Configure

I didn't include some dataset files because they take too much storage (some are ~1GB) Here is the step-by-step direction to set each of them up manually:

---

cifar10 Dataset

1. Open your notebook at 12-data/07-DUDL_data_data2colab.ipynb
2. Run the following 2 code cells:

# Download this on the shell if you haven't
!pip install torchvision

# FOR Image Processing
import torchvision
cdata = torchvision.datasets.CIFAR10(root='cifar10', download=True)
print(cdata)

3. Remove download=True from the second code cell after dataset is downloaded

---

emnist dataset

1. Use Google Collab (as this may not work on Jupyter Notebook!)
2. Open your notebook at 19-CNN/11-DUDL_CNN_EMNIST.ipynb
3. Run the following code cell:

cdata = torchvision.datasets.EMNIST(root='emnist',split='letters', download=True)

4. Remove download=True from the code cell after dataset is downloaded

---

fmnist dataset

1. Open your notebook at 20-CNNmilestone/DUDL_CNNmilestone_project3.ipynb
2. Run the following code cell:

trainset = torchvision.datasets.FashionMNIST(root='../fmnist', train=True, transform=transform, download=True)
devtest  = torchvision.datasets.FashionMNIST(root='../fmnist', train=False, transform=transform)

3. Remove download=True from the code cell after dataset is downloaded

---

stl10

1. Open your notebook at 21-transferlearning/03-DUDL_transfer_resnet.ipynb
2. Run the following code cell:

trainset = torchvision.datasets.STL10(root='./stl10', split='train', transform=transform, download=True)
testset  = torchvision.datasets.STL10(root='./stl10', split='test',  transform=transform)

3. Remove download=True from the code cell after dataset is downloaded