lecture 03
Convert trained model with Python and CoreML Tool
+
Style classification app
+
Python basics revisited

SeAts APp SEAtS ApP SEaTS APP

open anaconda
create an environment
with python 3.6
(what's this???)

select that new environment in anaconda
spyder disappears!

because this is a new environment!
install spyder

as usual, an AI-related project to wake us up

after today's lecture:
-- three levels of "readiness" of ML model
-- how to convert trained models from other formats into Apple's mlmodel 🥰
-- a deeper understanding of python 😉
(this is just our third python time!) 🫡

three stages of ML model

level 1: not ready at all
- newborn, needs training from scratch

level 2: trained, but on a different datasets/tasks
- needs fine-tuning (popular approach and we have seen this already)

level 3: trained, on the exactly same dataset and task
- everything is ready 🧐

Today we will be looking at the level 3, aka the most ready ML models. No training is needed.

so, can we just download the trained models and dra-n-drop them into the xcode project?
- nope

deep learning frameworks: caffe, pytorch, tensorflow, etc.

models trained using different frameworks have different formats and Apple can't just directly take these formats in.
Apple ML ecosystem has its own ML model format!

caffe: .caffemodel and .prototxt
pytorch: .pt
tensorflow: .tf2 and more
🍎: .mlmodel

like how you convert between audio/video formats
e.g.
.mp3 to .wav,
or .mp4 to .mov

we need to convert ML models into apple's .mlmodel

This part is handy when you want to plugin and play a trained model downloaded from github/huggingface, etc.

1. convert caffe model to .mlmodel

download this repo

caffe model zoo

import the model into style classification model and play

fun ai time part 1

2. convert pytorch model to .mlmodel

torch model zoo

recap from last lecture 01
How can we numerically represent audios and images in python?
-- There are libraries that read them into numbers...
How are these numbers being stored in python?
-- Numpy arrays !!! 💥💥💥

recap from last lecture 02
python
- for data pre-processing 🔨
-- os, makedirs
-- pandas, dataframe
- for holding numbers, arithmetic operations 🧮
-- numpy
- for reading audio 🔈
-- librosa
- for reading image 🎇
-- cv2
- for visualizing numbers👁
-- matplotlib

python note book basics 00

python note book basics 01
- variables
- lists
- conditionals
- loops
- functions

new series episode 2 queer the dataset

new series episode 2 jake elwes

python note book basics 02

fun ai time part 2

summary today
- readiness of ML models 💭
- create new python environment 💻
- convert caffe model into .mlmodel🍎
- style classification app🍎
- python basics enhanced 💻