Big

ML Two
Lecture 01
🤗Introduction to ML Two
+
👍First time training an AI model with CreateML: image classification

🎤As usual,
- a made-with-AI music video to wake us up,
- an AI research work that is relevant to that music video,
- more from this researcher Shiry Ginosar

In ML Two, we will:
- train our own AI models🌶️
- see example Apps that integrate a variety of AI models🌶️
- a variety of AI models: computer vision models (classification, object detection, etc.), image generation model, audio synthesis model, large language model, etc.👌

ML Two assessment:
- Part 2: A presentation around a bespoke ML model.
- start thinking about this "bespoke ML model" early!!!
- you can use your ML One presentation proposal as a starting point,
- of course you can also build something new!
- Make good use of CCI resources around, google, your peers, Mick and I 😎
- We also have a fantastic AI and Data Science technician Mayra Berrones 😎

question time! imagine you are the principal AI engineer at a fast-growing tech startup that is specialised in classifying fruit images being apple🍎 or banana🍌

what skills do you need? what are the steps to take to make a fruit image classification model ? (anything you can think of)

the hidden technical debt in machine learning system
Only a small fraction of real-world ML systems is composed of the ML code

1. DATA PREP
- - data collection
- - data pre-processing
2. TRAINING
- - fine tuning
- - from scratch
3. DEPLOYMENT
- - Apple dev, etc.

1. DATA PREP
- - data collection (p)
- - data pre-processing (p)
2. TRAINING
- - fine tuning (p,c)
- - from scratch (p,t)
3. DEPLOYMENT (c)
p: python 🐍
C: CoreML and CreateML 🍎🤖

Let's start building the fruit classification model🍎🍏!
We will be following this typical ML dev pipeline:
1. DATA PREP
- - data collection
- - data pre-processing
2. MODEL TRAINING
- - train a model in CreateML (super simple)
- - evaluate and improve
3. DEPLOYMENT

- download this starter project
- put it in a folder for ML Two stuff and unzip
- open *ICDemo-working xcodeproj*
- NOT the ICDemo-starter
- NOT the ICDemo-working mlproj

Step 2 data pre-processing:
- that's just raw data(images scattered in folders), often times we need to preprocess the raw data to our needs!
- this part is usually done by coding in python,
- but CreateML just makes our life easier 🤓

Step 2.1 Summon CreateML 😈
- open Xcode and select Xcode
→ Open Developer Tool
- → Create ML from the drop-down menu

Step 2 data pre-processing:
What data pre-processing do we need in this task?
- For supervised learning, we need data and their labels.
=>
- for image classification, we need images and their classes
=>
- How does CreateML know the corresponding label for each image?

🤗 Label info is inferred from file structure!
- this is also how we should prepare our img classification data (which is already done in this fruit classification example), when using CreateML
- Though when training in python, there are many other ways to feed in labels information...

In CreateML, we can firstly look at the training/evaluation tab,
where the higher accuracy the better

not working great 🫤
this is a good example of "you should not only looking at the accuracies,
but also inputting images to the model and seeing the classification result yourself!"

Let's bring this to our App!
- 1. export the model
- 2. drop the model file to the IOS App code.

😎Done! Here is a summary of what we have done so far:
- 1. Collect a dataset of images of different classes.
- 2. Organise the images into folders with folder name being the class name.
- 3. Open CreateML.
- 4. Drag and drop the images into the data pool.
- 5. Select the hyper-parameters to your liking and hit training!
- 6. Evaluate the results, adjust hyper-parameters, and train again!
- 7. Export the model and drop into the IOS App code.

💙Homework: find an image classification dataset with a NICE file structure (or build your own)
this one for example
train a classifier and import to App
send a demo