drguthals
commented
5 years ago Add a section called “What is Cognitive?”. This is a high-level description of models and their data, as context for the “easy button” Azure Cognitive APIs. Since these are smart kids we want them to know what they’re really looking at.
I like this idea, I think giving them some background/context is great.
Add a section called How does Cognitive Work?”. After the cool demos, get into the details of how a model is created, what machine learning is, how that factors into AI.
I think this is OK, just as long as it isn't too detailed, it's OK to stay a bit high level and hand-wavy at this stage since they won't actually be building the models themselves.
Add “Real World AI”. Show some case studies, not code, of real-world applications of cognitive.
Yes, I like this.
Do we need to do a quick GitHub primer?
Everything the students will see will be on Azure Notebooks, so we don't need to worry about GitHub.
The biggest weakness of most our code demos is that the output is long-form JSON. There are a couple of methods like “describe_image” that return human-readable results but many don’t. I’m wondering if we can’t provide a better JSON parser to give them more options for using the output, or if that idea will overcomplicate this project.
Yeah this isn't really nice. If you are able to create some simple Python code to better parse the JSON and have them copy that along with the code they are copying for making the call to the API, that would be great! (Or find something that already exists). Otherwise, I think we need to just have very clear examples of what the JSON looks like and what they are looking for and they can do pattern matching.
What are you using to create the markup? In the past I’ve used Visual Studio Code but it looks like it may be better to use that for test, heavy documents then interactive code docs use Notebook/Jupyter?
I've just been putting them in Azure Notebooks, which you can use for free here. That's what the students will be using also. If you want to just put them in this GitHub repo though, that is completely fine and I can move them over to Notebooks when we're ready 😄
If you can round this out, please don’t hesitate. In “Real World AI” case studies, what would be fun here, I wonder?
No problem 😄 I think these are fairly interesting! Maybe not the financial ones, but just knowing that AI is already pervasive in their world today is sometimes shocking to students. Smart cars, Alexa/Google Home/Siri/Cortana devices, video games.
danhermes
Questions:
Do we need to do a quick GitHub primer?
The biggest weakness of most our code demos is that the output is long-form JSON. There are a couple of methods like “describe_image” that return human-readable results but many don’t. I’m wondering if we can’t provide a better JSON parser to give them more options for using the output, or if that idea will overcomplicate this project.
What are you using to create the markup? In the past I’ve used Visual Studio Code but it looks like it may be better to use that for test, heavy documents then interactive code docs use Notebook/Jupyter?
BTW I think I’m probably stronger on the serious coolness factor of these technologies than on lightness and fun. I looked at the Pokemon example and that was awesome but I would never think of that (not to mention the time crunch). If you can round this out, please don’t hesitate. In “Real World AI” case studies, what would be fun here, I wonder?