4 ms·

MadeWithML – A practical approach to learning production machine learning

175 points by lineardigress 5y ago

lineardigress 5y ago
There is a website for the lessons too: https://madewithml.com/ https://madewithml.com/ Disclosure: I'm a contributor in their open-source community activities and also in the ML space. Feel open to reach out to me if you have any questions.
- roflchoppa 5y ago
  That’s one way to self promote. Cool repo.
- wunderwuzzi23 5y ago
  Good stuff, at high level it seems security engineering aspects might be missing. I put together content around threat modeling, infrastructure attacks as well as ML specific attacks with practical examples (like using adversarial robustness toolbox from Linux foundation). Details here: https://github.com/wunderwuzzi23/mlattacks https://github.com/wunderwuzzi23/mlattacks
  - raghavbali 5y ago
    Something I have been looking for quite some time. Thank you for this.
- lineardigress 5y ago
  Personally there are many ways to implement machine learning in production and this resource does a good job on giving you a detailed grand tour of all the fundamentals you need to think about. And to make the concepts more concrete, each unit has code implementation for an end-to-end project. While some lessons can stand to be more well connected to the end-to-end project, the sheer amount of detail on some of the lessons simply don't exist anywhere else. The testing (https://madewithml.com/courses/mlops/testing/ https://madewithml.com/courses/mlops/testing/) and monitoring (https://madewithml.com/courses/mlops/monitoring/ https://madewithml.com/courses/mlops/monitoring/) lessons are clear examples of this.
seibelj 5y ago
> How come there's a lot of buzz around ML but we grossly fall short when it comes to deployed ML in real world applications (especially in certain industries). There are many reasons for this, including but not limited to: > data: not enough of the right data. > cost: the data, compute, storage and talent resources needed for ML allow mostly large companies to benefit from it. > utility: most ML content you see online are gimmicky tasks that can't extend into valuable real world applications. > trust: ML needs to be engineered reliably and robustly. This is a very honest take on ML / AI. I work in blockchain and what drives me crazy is that crypto / blockchain has delivered huge, tangible successes and value (measured easily and accurately via real-time prices) but the HN cognoscenti orgasms over ML and considers blockchain a scam. IMO the situation is flipped - crypto is proving its success in a massive, revolutionary way while ML / AI is pretty stagnant at delivering world-changing products. It's cool that bots are sorting and picking at the warehouse but the value has been overstated by many orders of magnitude (thus far). ML is a tool in a box right now, one of many that can potentially solve a class of problems, but it isn't (yet) a paradigm shift in how businesses operate.
- HWR_14 5y ago
  I'll ask the question I seem to always ask: What is a problem that is solved by blockchain that cannot be solved by just having a central authority?
  - seibelj 5y ago
    Blockchains are value-transfer networks. The value transfer layer has historically been abused as a rule rather than the exception: - Rent seeking, such as by Visa and Mastercard that continually try and raise merchant fees - Censorship, such as with OnlyFans and its attempted blacklisting by the banking system - As a weapon of war, as with the treasury blocking entities via their monopoly of the global reserve currency and their licensing / penalty system of banks regardless of their country - Debasement of money in general by the money printers Blockchains, when properly implemented as closely as can be attempted using decentralized engineering techniques, are neutral plumbing for value transfer. Ethereum is owned by no one, thus competitors can use it, and fees can be optionally added by middlemen if their services justify the fees. The benefit to the world is choice in the world of finance, something society has never had until recently.
chairmanwow1 5y ago
The quality of the attached site makes me think this guy is making a good amount of money by operating this site. Anyone have any insight? I can’t seem to crack the code.
- amrrs 5y ago
  His course content is free, I'm not sure if he's making any money out of the course.
- mellavora 5y ago
  Well, according to the website, it is a form of philanthropy (see quote below). Also, it sure gives the author a lot of credibility in the field, which could lead to interesting future career opportunities. I don't see anything nefarious. Seems to be a pretty straight-up effort to improve the world. He says these are his reasons: While this content is for everyone, it's especially targeted towards people who don't have as much opportunity to learn. I believe that creativity and intelligence are randomly distributed while opportunities are siloed. I want to enable more people to create and contribute to innovation.
- matsemann 5y ago
  You made me expect more. Not to be dismissive of the effort involved, but I was hoping for a complete course. The stuff I checked out looked good, but not very comprehensive. It's some code examples for each category and some links to other resources. For instance "data labeling" could be a book in itself but is a few paragraphs long. Dropout is mentioned in a few sentences as "a great technique". It's good to get an overview over the various things one needs to learn. But I cannot imagine being ready to put anything in production after reading through this, so no MLOps exactly.

175 points by lineardigress 5y ago

There is a website for the lessons too: https://madewithml.com/ https://madewithml.com/ Disclosure: I'm a contributor in their open-source community activities and also in the ML space. Feel open to reach out to me if you have any questions.

roflchoppa 5y ago
That’s one way to self promote. Cool repo.
wunderwuzzi23 5y ago
Good stuff, at high level it seems security engineering aspects might be missing. I put together content around threat modeling, infrastructure attacks as well as ML specific attacks with practical examples (like using adversarial robustness toolbox from Linux foundation). Details here: https://github.com/wunderwuzzi23/mlattacks https://github.com/wunderwuzzi23/mlattacks
- raghavbali 5y ago
  Something I have been looking for quite some time. Thank you for this.
lineardigress 5y ago
Personally there are many ways to implement machine learning in production and this resource does a good job on giving you a detailed grand tour of all the fundamentals you need to think about. And to make the concepts more concrete, each unit has code implementation for an end-to-end project. While some lessons can stand to be more well connected to the end-to-end project, the sheer amount of detail on some of the lessons simply don't exist anywhere else. The testing (https://madewithml.com/courses/mlops/testing/ https://madewithml.com/courses/mlops/testing/) and monitoring (https://madewithml.com/courses/mlops/monitoring/ https://madewithml.com/courses/mlops/monitoring/) lessons are clear examples of this.

seibelj 5y ago

> How come there's a lot of buzz around ML but we grossly fall short when it comes to deployed ML in real world applications (especially in certain industries). There are many reasons for this, including but not limited to: > data: not enough of the right data. > cost: the data, compute, storage and talent resources needed for ML allow mostly large companies to benefit from it. > utility: most ML content you see online are gimmicky tasks that can't extend into valuable real world applications. > trust: ML needs to be engineered reliably and robustly. This is a very honest take on ML / AI. I work in blockchain and what drives me crazy is that crypto / blockchain has delivered huge, tangible successes and value (measured easily and accurately via real-time prices) but the HN cognoscenti orgasms over ML and considers blockchain a scam. IMO the situation is flipped - crypto is proving its success in a massive, revolutionary way while ML / AI is pretty stagnant at delivering world-changing products. It's cool that bots are sorting and picking at the warehouse but the value has been overstated by many orders of magnitude (thus far). ML is a tool in a box right now, one of many that can potentially solve a class of problems, but it isn't (yet) a paradigm shift in how businesses operate.

HWR_14 5y ago
I'll ask the question I seem to always ask: What is a problem that is solved by blockchain that cannot be solved by just having a central authority?
- seibelj 5y ago
  Blockchains are value-transfer networks. The value transfer layer has historically been abused as a rule rather than the exception: - Rent seeking, such as by Visa and Mastercard that continually try and raise merchant fees - Censorship, such as with OnlyFans and its attempted blacklisting by the banking system - As a weapon of war, as with the treasury blocking entities via their monopoly of the global reserve currency and their licensing / penalty system of banks regardless of their country - Debasement of money in general by the money printers Blockchains, when properly implemented as closely as can be attempted using decentralized engineering techniques, are neutral plumbing for value transfer. Ethereum is owned by no one, thus competitors can use it, and fees can be optionally added by middlemen if their services justify the fees. The benefit to the world is choice in the world of finance, something society has never had until recently.

chairmanwow1 5y ago

The quality of the attached site makes me think this guy is making a good amount of money by operating this site. Anyone have any insight? I can’t seem to crack the code.

amrrs 5y ago
His course content is free, I'm not sure if he's making any money out of the course.
mellavora 5y ago
Well, according to the website, it is a form of philanthropy (see quote below). Also, it sure gives the author a lot of credibility in the field, which could lead to interesting future career opportunities. I don't see anything nefarious. Seems to be a pretty straight-up effort to improve the world. He says these are his reasons: While this content is for everyone, it's especially targeted towards people who don't have as much opportunity to learn. I believe that creativity and intelligence are randomly distributed while opportunities are siloed. I want to enable more people to create and contribute to innovation.
matsemann 5y ago
You made me expect more. Not to be dismissive of the effort involved, but I was hoping for a complete course. The stuff I checked out looked good, but not very comprehensive. It's some code examples for each category and some links to other resources. For instance "data labeling" could be a book in itself but is a few paragraphs long. Dropout is mentioned in a few sentences as "a great technique". It's good to get an overview over the various things one needs to learn. But I cannot imagine being ready to put anything in production after reading through this, so no MLOps exactly.