[Day 134] Finished CS109 + Scottish dataset project + Started MLOps zoomcamp by DataTalks club

 Hello :)
Today is Day 134!


A quick summary of today:
  • covered the last 2 lectures of Stanford's CS109
  • processed 4 more videos for the Scottish accent dataset
  • started MLOps zoomcamp by DataTalks club
  • watched a nice video comparing the roles of data scientist vs AI engineer


Firstly, on lecture 27: advanced probability and lecture 28: future of probability from Stanford's CS109

What a course! Sad it is over. Chris Piech - what an extraordinary professor! 

Both lectures were more or less an overview of the course and professor Piech's hopes for the future of probability and his hopes for his students.

Lecture 27

Around 2013, autograd was introduced for the first time and this allowed for backprop to be automated, instead of manual calculations. This was one of the reasons for deep learning to explode.

The professor also talked a bit about diffusion models

#1 remove the noise
#2 predict where the noise is

And a bit about language models (at the time of recording ChatGPT had just come out). And he introduced a basic version of RNN/LSTM. 

Lecture 28
Besides an overview of what was covered throughout the previous 27 lectures, he talked about potential usage of AI in education and how it can help provide feedback to students, and in medicine. 

Secondly, about the Scottish dataset project

Today I cut videos 'Scottish phrases 5-8' (4 videos). And now the total time of the audio clips we have in the dataset is 10 minutes :party: We are improving slowly. 


Thirdly, started MLOps zoomcamp by DataTalks club

Last night was the opening day for the MLOps camp that I signed up for. It is one (if not the) most popular open courses for MLOps, so I was excited to start it. Well today I covered module 1 and its homework. Below are notes from the intro lecture + module 1 + the homework. 

Intro lecture

The course will cover the following topics

Module 1

I think the coolest thing was learning about the different MLOps maturity levels (introduced by Microsoft)

Level 0: no MLOps

  • no automation
  • all code in jupyter
  • good for experiments

Level 1: DevOps, No MLOps

  • there are experienced devs helping the data scientists
  • some automation
  • releases are automeated
  • unit & integration tests
  • CI/CD
  • ops metrics
  • no experiment tracking
  • no reproducability
  • data scientists separated from engineers

Level 2: Automated training

  • training pipeline
  • experiment tracking
  • model registry
  • low friction deployment
  • data scientists work with engineers
  • good level if we have 2-3 ML cases

Level 3: Automated deployment

  • easy to deploy models
  • pipeline is: data prep, model traning, model deployment 
  • A/B tests between models
  • some model monitoring

Level 4: Full MLOps automation

  • automatic training
  • automatic retraining
  • automatic deployment
  • A/B tests
  • approaching a zero-downtime system

Not all orgs need to be on level 4. Level 3 is still fine because we can have a human making that final decision whether a model goes live. So we need to judge what level is best for a particular project.

The homework

I uploaded all my work to my github.

The homework followed the module 1 materia on creating a linear regression model that predicts the taxi travel time from point A to point B. It involves some data preprocessing, model creation and exporting the model with pickle. 

Q1: loading the data and how many columns there are


Q2: What's the standard deviation of the trips duration in January?


Q3: What fraction of the records left after you dropped the outliers?


Q4: What's the dimensionality of the matrix (number of columns) after using DictVectorizer?

Q5: What's the RMSE on train?

We can interpret this as being 7.6 minutes wrong on average in our ride duration predictions.

Q6: What's the RMSE on validation?

After loading the val dataset (data from February 2023, train is January 2023), I got the result

Overall, pretty satisfied, and excited for the future modules and homeworks. 


Finally, the comparison between a data scientist and an AI engineer

This IBM video was in my recommended on youtube so I clicked on it to see. And the below pic is the final explanation.

Some abbreviations: FM: foundation model, FE: feature eng, CV: cross-val, HPT: hyperparam tuning, PEFT: param efficiant finetuning. Of course the below is not static, DS might work on prescriptive cases, and AI eng can work with structured data.



That is all for today!

See you tomorrow :)

Popular posts from this blog

[Day 198] Transactions Data Streaming Pipeline Porject [v1 completed]

Image
 Hello :) Today is Day 198! A quick summary of today: data streaming pipeline project [v1 done] Here is a link to the project's repo. Well ... I did not know I can do it in a day (~14 hours) after yesterday's issues but here we are. Turns out in order to insert the full (~70 variables with nested/list structure), I need the proper pyspark schema. And yesterday I did not have that and that is why when I was reading data in the kafka producer I was getting NULL in the columns - my schema was wrong. Well today I not only fixed the schema for the 4 variables I had yesterday, but included *all* the variables that come from the Stripe API ~ 70 (for completeness).  When I run docker-compose, the data streams and is input into the postgres db (and is still running). Unfortunately, the free Stripe API for creating realistic transactions has a limit of 25, so every 3 seconds, 25 new transactions are sent to the db. It has been running half the day (since I got that set up) and as I am wr
Read more

[미리 공부] 기초 통계 복습 (Day 1는 1월2일)

Image
모험을 시작하기 전에 기초 지식을 복습하고자 했다. 오늘 SPSS 대안 프로그램을 찾아보려고 해서 JASP에 대해 알게 되었다. 유용한 프로그램인 것 같아서 선회귀와 기술통계를 내려고 했는데 재미있었다. 그런데 JASP에 대해 더 알기 전에 '기초 통계 지식을 좀 복습을 하고자 하면 좋을 것 같아'란 생각을 들었다. 다행히, Coursera에서 Stanford University의 Guenther Walther 교수님께서 진행된  Introduction to Statistics  무료 강좌가 있다.  좀 부족한 부분은 다양한 검정통계 하는 거고 (F test, t-test, chi-square 등) 이제 JASP 아니면 다른 통계 프로그램 사용하게 되어도 이런 부분을 좀 더 자세히 집중하여 공부하면 된다.  특히 homoscedasticity 및 heteroscedasticity 개념을 기억에 남았다. Homoscedasticity (선): Definition: In a homoscedastic dataset, the variance of the errors (residuals) is constant across all levels of the independent variable(s). In simpler terms, the spread of the residuals is the same throughout the range of predictor values. Heteroscedasticity (악): Definition: Heteroscedasticity occurs when the variance of the errors is not constant across all levels of the independent variable(s). In other words, the spread of residuals changes as the values of the independent variable(s) change. (Udemy 사진) 강의
Read more

[Day 61] Stanford CS224N (NLP with DL): Machine translation, seq2seq + a side CDCGAN mini project

Image
 Hello :) Today is Day 61! A quick summary of today: Covered Lecture 7 : machine translation, seq2seq, attention from Stanford CS224N Tried to make a conditional DCGAN to generate MNIST numbers ( colab ) ( kaggle ) I will first cover the GAN story (then will share my notes from the lecture) So... while watching and taking notes today, I started thinking, what if I can use my notes as data to a model and afterwards, when I want, I can give it raw string text and it will output text in the format of my notes (with my handwriting). Well I started looking around and actually the first model architecture that came to my mind was the GAN (specifically conditional GAN) - I remembered there was a GAN architecture that alongside the pictures, we can give it the labels, and then on-demand generate. In retrospect, there are of course others, but I decided to go with GAN.  For maybe 2 hours I busted my head trying to make a simple model with the EMNIST dataset (english characters), and I kept gett
Read more