My new TensorFlow for Deep Learning course is now live on Udemy
Done a beginner-level ML course? Ready to step up and learn deep learning? Like writing lots of code? This course is for you.
Friends, I’m writing to you from Brisbane’s Northern Beaches. It’s raining here. 4 pm feels like 6 pm.
During lockdowns of 2020, I got TensorFlow Developer Certified. And I made a video about it which people seemed to enjoy.
A lot of questions came in asking how I did it.
I answered with “I read the Hands-on Machine Learning Book end-to-end and then went through the practice questions for the exam.”
Which is true. And you could do the same steps if you wanted to pass the exam. Or you could take my new TensorFlow for Deep Learning course (now on Udemy, use this link for an excellent valued launch deal).
You can also sign up directly on the Zero to Mastery Academy (doing so here will give you access to all ZTM courses).
Specifically, the Zero to Mastery TensorFlow for Deep Learning course will teach you:
Fundamental concepts of deep learning (we teach concepts using code)
How to write TensorFlow code (lots of it)
How to pass the TensorFlow Developer Certification (optional, but requires 1 & 2)
I’m so excited for this launch and am stoked that people are loving it so far.
Certifications are *nice* to have not *need* to have
I got asked whether the certification is necessary in a recent livestream Q&A about the course.
It’s not.
In the tech field, no certification is needed. If you have skills and demonstrate those skills (through a blog, projects of your own, a nice-looking GitHub), that is a certification in itself.
A certification is only one form of proof of skill.
Rather than an official certification, I’m a big fan of starting the job before you have it. For example, if you have some ideal role you’d like to work for a company as. Say, a machine learning engineer. Use your research skills to figure out what a machine learning engineer would do day to day and then start doing those things.
Use courses and certifications as foundational knowledge then use your own projects to build specific knowledge (knowledge that can’t be taught).
How is the course taught?
It’s taught code-first, code-along, apprenticeship style. This means in every video where a new concept is being introduced, it gets introduced using code.
Who should take the course?
This is a beginner-level course.
If you’ve got 1+ years of experience with deep learning/TensorFlow/PyTorch (PyTorch is another deep learning framework like TensorFlow), you shouldn’t take it, use your skills to make something instead.
This course is for coders with ~6-months of experience writing Python code who want to learn about deep learning and how to build neural networks for various problems using TensorFlow.
What are the prerequisites?
Even though it’s for beginners, there are some prerequisites I’d recommend.
6+ months writing Python code. Can you write a Python function that accepts and uses parameters? That’s good enough. If you don’t know what that means, spend another month or two writing Python code and then come back here.
At least one beginner machine learning course. Are you familiar with the idea of training, validation and test sets? Do you know what supervised learning is? Have you used pandas, NumPy or Matplotlib before? If no for any of these, I’d go through at least one machine learning course that teaches these first and then coming back.
Comfortable using Google Colab/Jupyter Notebooks. This course uses Google Colab throughout. If you have never used Google Colab before, it works very similar to Jupyter Notebooks with a few extra features. If you’re not familiar with Google Colab notebooks, I’d suggest going through the Introduction to Google Colab notebook.
Plug: The Zero to Mastery beginner-friendly machine learning course (I also teach this) teaches all of the above (and this course is designed as a follow-on).
With this being said, you can watch the first 14-hours (see notebooks 00, 01, 02 below) of the course on YouTube. If you’re interested, I’d encourage you to try it out and see how you go.
What’s in the course?
You can see all of the materials on the course GitHub (this is the source of truth for the course, all major announcements/updates/Q&A will go here as well).
The course is broken down into 11 notebooks (starting from 0), all of which cover something specific about TensorFlow or deep learning.
Notebook 0 — TensorFlow Fundamentals
Introduction to tensors (creating tensors with TensorFlow)
Getting information from tensors (tensor attributes)
Manipulating tensors (tensor operations)
Tensors and NumPy
Using @tf.function (a way to speed up your regular Python functions)
Using GPUs with TensorFlow
Key takeaways:
Learn how to create and manipulate the fundamental building block of deep learning: the tensor
Learn to represent data as numbers (so it can be used with a machine learning model)
1 — Neural Network Regression with TensorFlow
The architecture of a neural network regression model
Input shapes and output shapes of a regression model
X: features/data (inputs)y: labels (outputs)
Creating custom data to view and fit
Steps in modelling with TensorFlow
Creating a model
Compiling a model
Defining a loss function
Setting up an optimizer
Creating evaluation metrics
Fitting a model (getting it to find patterns in our data)
Evaluating a regression model
Visualizing the model ("visualize, visualize, visualize")
Looking at training curves
Compare predictions to ground truth (using regression evaluation metrics)
MAE — mean average error
MSE — mean square error
Saving a model (so we can use it later)
Loading a model
Key Takeaways:
Build TensorFlow sequential models with multiple layers
Prepare data for use with a machine learning model
Learn the different components which make up a deep learning model (loss function, architecture, optimization function)
Learn how to diagnose a regression problem (predicting a number) and build a neural network for it
2 — Neural Network Classification with TensorFlow
The architecture of a neural network classification model
Input shapes and output shapes of a classification model
X: features/data (inputs)y: labels (outputs)"What class do the inputs belong to?"
Creating custom data to view and fit
Steps in modelling for binary and multi-class classification
Creating a model
Compiling a model
Defining a loss function
Setting up an optimizer
Finding the best learning rate (a value that decides how fast a model learns)
Creating evaluation metrics
Fitting a model (getting it to find patterns in our data)
Improving a model
The power of non-linearity
Understand different classification evaluation methods
Visualizing the model ("visualize, visualize, visualize")
Looking at training curves
Compare predictions to ground truth (using classification evaluation metrics)
Precision
Accuracy
Recall
F1-score
Key takeaways:
Learn how to diagnose a classification problem (predicting whether something is one thing or another)
Build, compile & train machine learning classification models using TensorFlow
Build and train models for binary and multi-class classification
Plot modelling performance metrics against each other
Match input (training data shape) and output shapes (prediction data target)
3 — Computer Vision and Convolutional Neural Networks with TensorFlow
Getting an image dataset to work with (computer vision models need images of some kind)
Architecture of a convolutional neural network (CNN)
A quick end-to-end example (what we're working towards)
Steps in modelling for binary image classification with CNNs
Becoming one with the data (inspecting and understanding your data)
Preparing data for modelling (using ImageDataGenerator)
Creating a CNN model (starting with a baseline)
Fitting a model (getting it to find patterns in our data)
Evaluating a model
Improving a model
Making a prediction with a trained model
Steps in modelling for multi-class image classification with CNNs
Same as above (but this time with a different dataset)
Key takeaways:
Build convolutional neural networks with Conv2D and pooling layers
Learn how to diagnose different kinds of computer vision problems
Learn how to build computer vision neural networks
Learn how to use real-world images with your computer vision models
4 — Transfer Learning with TensorFlow Part 1: Feature Extraction
Introduce transfer learning (leveraging what one model has learned somewhere else and how to adjust it to our own problem, a way to beat all of our old self-built models)
Using a smaller dataset to experiment faster (10% of training samples of 10 classes of food)
Build a transfer learning feature extraction model using TensorFlow Hub
Introduce the TensorBoard callback to track model training results
Compare model results using TensorBoard
Key takeaways:
Learn how to use pre-trained models to extract features from your own data
Learn how to use TensorFlow Hub for pre-trained models
Learn how to use TensorBoard to compare the performance of several different models
5 — Transfer Learning with TensorFlow Part 2: Fine-tuning
Introduce fine-tuning, a type of transfer learning to modify a pre-trained model to be more suited to your data
Using the Keras Functional API (a different way to build models in Keras)
Using a smaller dataset to experiment faster (10% of training samples of 10 classes of food)
Data augmentation (how to make your training dataset more diverse without adding more data)
Running a series of modelling experiments on our Food Vision data
Model 0: a transfer learning model using the Keras Functional API
Model 1: a feature extraction transfer learning model on 1% of the data with data augmentation
Model 2: a feature extraction transfer learning model on 10% of the data with data augmentation
Model 3: a fine-tuned transfer learning model on 10% of the data
Model 4: a fine-tuned transfer learning model on 100% of the data
Introduce the ModelCheckpoint callback to save intermediate training results
Compare model experiments results using TensorBoard
Key takeaways:
Learn how to set up and run several machine learning experiments
Learn how to use data augmentation to increase the diversity of your training data
Learn how to fine-tune a pre-trained model to your own custom problem
Learn how to use Callbacks to add functionality to your model during training
6 — Transfer Learning with TensorFlow Part 3: Scaling Up (Food Vision mini)
Downloading and preparing 10% of the Food101 data (10% of training data)
Training a feature extraction transfer learning model on 10% of the Food101 training data
Fine-tuning our feature extraction model
Saving and loading a trained model
Evaluating the performance of our Food Vision mini model trained on 10% of the training data
Making predictions with our Food Vision mini model on custom images of food
Finding the most wrong predictions
Key takeaways:
Learn how to scale up an existing model
Learn how to evaluate your machine learning models by finding the most wrong predictions
Beat the original Food101 paper using only 10% of the data
7 — Milestone Project 1: Food Vision 🍔👁
Combine everything you've learned in the previous 6 notebooks to build Food Vision: a computer vision model able to classify images of 101 different kinds of foods
Our model well and truly beats the original Food101 paper
8 — NLP Fundamentals in TensorFlow
Learn to diagnose a natural language processing problem
Key takeaways:
Learn to:
Preprocess natural language text to be used with a neural network
Create word embeddings (numerical representations of text) with TensorFlow
Build neural networks capable of binary and multi-class classification using:
RNNs (recurrent neural networks)
LSTMs (long short-term memory cells)
GRUs (gated recurrent units)
CNNs
Learn how to evaluate your NLP models
9 — Milestone Project 2: SkimLit 📄🔥
Replicate the model which powers the PubMed 200k paper to classify different sequences in PubMed medical abstracts (which can help researchers read through medical abstracts faster)
10 — Time Series fundamentals in TensorFlow & Milestone Project 3: BitPredict (coming soon)
Learn how to diagnose a time series problem (building a model to make predictions based on data across time, e.g. predicting the stock price of AAPL tomorrow)
Prepare data for time series neural networks (features and labels)
Understanding and using different time series evaluation methods
Build time series forecasting models with TensorFlow
Preparing for the certification
As I said before, the certification is optional. And after going through the above sections, you’ll well and truly have the skills to prepare and take on the TensorFlow Developer Certification yourself.
This section will consist of pointing you to various resources to further prepare yourself.
Namely, the TensorFlow Developer Certification Handbook. You should read this and go through each requirement guideline and see if you can replicate it with TensorFlow code (if you can’t, you’ll now have the skills to figure out how).
Get ready to dream in tensors
There’s a celebratory deal going on the Udemy version of the course for the next day or so.
Use the code TF2021LAUNCH to get it (the link above should have it already applied), the code will also work on my beginner-friendly Machine Learning and Data Science course.
Otherwise, you can sign up on the Zero to Mastery Academy and get access to both (as well as all other ZTM courses).
If you have any more questions, I did a live Q&A on YouTube discussing the launch:
Otherwise, leave a note on the GitHub discussions page, that way others can see it too.
Big dogs gotta eat,
Daniel