Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: Graph-based notebook for data scientists and researchers (amie.ai)
171 points by hsuominen 5 months ago | hide | past | web | favorite | 53 comments

The live demo is a really neat detail, but it can be overwhelming for an uninitiated user.

I first didn't even notice it was a live demo until my mouse was over it and I wanted to scroll down and instead the graph in the demo zoomed out.

Some feedback:

- Just today on HN: https://news.ycombinator.com/item?id=18410628

- I would not do a live demo but instead illustrate using pictures and perhaps a little intro video.

- It's not made clear enough what the core differentiators of the product are vs. existing process tools, such as Trello or Asana. In fact I currently don't see the benefits of a graph-based approach over other tools. Doesn't mean that there aren't any, just that they aren't made clear.

- In the live demo: the box being displayed when I hover over a tree node is almost always (partly) off screen.

I like having a live demo available, but not on the landing page. It should be one click away to avoid this kind of confusion.

Noted, thanks!

I think the live demo is awesome! I think you should keep it.

But: it also took me a while to see that it is interactive.

There should be a big label "Live Demo" above it (the small text on the right is easy to miss)

(Unfortunately, for me, all the live demo does is show that the product is still pretty rough.)

Thanks! It is, of course, new, but also less rough when you use the actual product and not the live demo.

Thank you for your feedback! The main advantages of the graph are: 1. you can get an easier overview of complex processes and how they connect. 2. The tracked parameters are inherited. So if you are tracking parameters that don't change to the next iteration, you will still have them stored in the process. It also remembers how they connect, so if you change a name upstream it will cascade down. 3. You can plot the leaves according to their parameters. If you click on the plotting button in the top left and choose model_name vs. mae for example, you can compare how different models performed and still see how their iterations connect.

The "need more pixels" message is quite frustrating. I get that it might be not practical to support small resolutions, but I was unable to see the demo even on "desktop mode" in landscape on a really big phone. How many pixels do you need? If it's still not practical to support that, a demo video player would be a great fallback.

You are absolutely right, we shall soon make a video for that scenario. For now we just bail if the device has less than 400x600 pixels "big" viewport, and if the "zoom" property is not supported. The demo would actually work fine but we feel the ux becomes too awkward :/

Great concept. It seems like the graph view is begging to convey more information. At the very least it seems appropriate for the nodes to have an icon based on what kind of data they contain, for instance, a paperclip or document for an attached paper, a histogram for univariate data, a regression plot for bivariate, etc.

Noted, thank you for your feedback!

If anybody is curious, CTO of amie here, we run on elm and a mix of Haskell and go, our db of choice is of course postgres !

Could you give more info on how this is better (or when this is better) than an ordinary notebook like evernote. I read through the homepage and found this:

> which lets you directly model your workflow and document your reasoning without the mental overhead of a folder structure.

but honestly i still don't understand what the main value prop is.


Of course, thank you for asking. An ordinary notebook only lets you order notes linearly. As soon as you try out different paths and switch back and forth between them, for example, different models, you can lose the overview. With the graph, for any result, you can always go back an understand exactly where it came from. We make this extra easy by letting you plot the notes according to the parameters they are tagged with. You can try it in the live demo by clicking on the graph button on the top left and select "model_name" and "mae" to compare how the models performed. Besides this (compared to Evernote or a paper notebook), we also have a python client that lets you integrate documentation directly into taking data from equipment or data science.

Is this intended to be a notebook in the jupyter sense or in the evernote sense? Initially I thought the former (and so pretty much discounted the possibility of using it as a personal knowledge base) but your response made me wonder if I'm missing something

I've tried pretty much everything on the planet looking for a good knowledge base/personal wiki that is graph based. My current tool of choice is a humble little app called Scapple, but it's far from satisfactory.

Both and neither ? :P

We think knowledge is much complex than a simple linear note taking process (Evernote), but also needs (at times) simpler tool than a full fledged computing platform (Jupyter).

With amie one can enrich ones notes with whatever metadata one wants to track, we call them key value pairs, branch off whenever one is exploring parallel ideas and connect every piece of ones knowledge via references.

One can do it all via a Gui, just like in Evernote, but also via our python client giving the user full flexibility.

While we have a lot of visual features coming to make the note taking part neater, we won't support a fully featured markdown-like input for text entry in the near future tho.

Let us know if you can use it to build your personal knowledge base, we think you're going to love amie!

Have you considered adding version control to this?

Yes indeed! Already now, under the hood, everything is version controlled :D We plan to expose a much more fine grained access to version control with a focus on the Jupiter notebook. We also have in the pipeline a neat (we think) visual way to see the diffs in the knowledge one builds using our graph.

For some reason I have added intercom.io on my NoScript in the past. I DuckDuckGo-ed it and it returns: "Intercom is a priceless tool that helps us drive Sales and Support through every step of the funnel."

I now will add sentry.io in my AdBlock+ list.

I am getting a login in screen with nothing else. I value my privacy over your service. But I sincerely wish you the best on your endeavors.

Intercom is just a chat widget (and backend system) for customer support / sales, and Sentry is a kind of hosted error log. You can read both of their privacy policies.



Also if you're going to be a privacy zealot you might as well switch to uBlock Origin over Adblock+.

Is a chatbot necessary for your product to work? Is it possible to see and test your product without these adins? Seems odd that a blank screen is what happens when your chat widget / hosted error log doesn’t work.

It's not (necessarily) a chat bot. I worked at a company that used Intercom, there were actually customer support reps responding to inquiries. It was actually a nice platform for triaging inquiries and support requests. You could even have side-conversations among employees in it, e.g. when one of our models gave weird predictions they would tag me in the conversation so I could tell them how to respond to the customer.

That said, having your website break when it's not loaded is bad.

No, if you have cookies disabled it can't log on to the live demo. As mentioned before, it's not a chatbot, it is just a way for you to get in touch with us right out of the app without opening an email client. If we are around we will also answer in the chat directly.

Thanks! Sorry to hear, but we need those tools. We are an EU company and thereby regulated by GDPR, so we are obligated to treat your data very responsibly. You can learn more about the product (without intercom and sentry) on our blog: https://medium.com/amie-ai/amie-use-case-data-driven-winemak... Feel free to get in touch!

Not GP, but this is exactly the reply that I was hoping to hear. The GDPR has done more for human rights for the masses than any other initiative in the past 70 years.

>The GDPR has done more for human rights for the masses than any other initiative in the past 70 years.

Talk about being egocentric. The world has larger parts that don't even have access to Internet.

Internet access shouldn't be a threshold issue when it comes to data privacy rights.

Facebook, for example, likely has a fair amount of personal data about a lot of people who don't have meaningful internet access. Something as simple as their contact information being saved in the phone of someone who uses Facebook mobile is enough for the company to create a "shadow profile" for them. That kind of data collection doesn't benefit the data owner by providing a "free" service in exchange for their information and it occurs entirely without their knowledge or consent. This is exactly the kind of privacy violation GDPR seeks to curtail and I think the benefits can and do extend to the global masses more than most people realize.

And also forced many websites to open with a popup 'Do you want to accept cookies?'. To the uninformed, could be a very mysterious message.

Cookie warnings are not GDPR but previous legislation.

I think the Rust Code of Conduct might beat GDPR

...Black people couldn't vote in the US 70 years ago.

> ...Black people couldn't vote in the US 70 years ago.

Black people have been guaranteed equal voting rights in theory since the 15th amendment was ratified in 1870, which was 148 years ago. (And free black citizens existed and could vote in parts of the USA before that.)

Black people since then have faced practical barriers to equal voting in many parts of the country, continuing to the present day.

I'd say marriage equality goes deeper than GDPR.


I found it hard to figure out where this could fit in my job as a researcher. I mean even a potential use case. I was ready to give up, but since you are EU based I'll try harder now.

It's not you, it's Intercom and Sentry that they don't trust.

I really don't understand why you are blocking Sentry, it's for error/exception logging and makes the developers behind the product so much faster at fixing stange issues that you might run into.

The layout looks weird in firefox, did you only test with chrome?

Yes, only Chrome and Safari are fully supported right now. It should tell you actually. Thanks for asking!

I've been working on a similar project for a bit now, really think this is a good idea and good job for shipping!

Thanks! Feel free to get in touch, maybe we can exchange ideas.

Just random nitpicks: The "interact with me" emoji is hard to make out, looks like a glitch at first. Emojis should be images, for both consistency and making sure people don't get boxy question marks if their browser can't render. "toggle fullscreen" is off-center vertically. The demo frame is messed up in Edge. The text in the about page needs a max width. Like others mentioned the demo is a bit overwhelming, it also has some minor UX issues and is generally the weakest link design wise at the moment.

Why do I need to allow your cookie just to view your landing page?

The page contains a live demo, that needs a cookie.

But you can't tell that without allowing the cookie - you just get a login prompt and 'not allowed to view this resource'.

Thanks, yes, we have to fix that.

Not very well I'm afraid. Firefox/uBlock/noScript just results in a blank page.

Supposedly this is for any device, but it doesn't work well on my cell phone. The interactive demo says "not enough pixels" and the chat widget gets in the way.

For smaller screens, I think it's better not to try to put as much stuff on the front page? A button going to another screen might work better than a widget.

I would have loved this during my post-doc!

you can always do another post-doc :P

Jokes aside, if you have time to get in touch with us we'd love to chat about the experience you had with the tools (or the lack thereof ) during your post doc!

I've been waiting for a graph view for Jupyter since it was called IPython. This is great!

You are going to love the next update which will bring full integration with Jupyter notebook and Jupyter lab! Keep an eye open!

Apologize, but I'm super confused about this product. How is this compared to say codalab (https://worksheets.codalab.org/) for organizing experiments?

No need to apologise :D It is very different in our view because we build a graph that connects all experiments you do. This allows you to get an overview of your entire discovery process. In our approach users can use the full power of a programmatic access, via our python client, and the full power of a Gui. With amie one can see for example: which feature performed best using a model Y after a grid search that resulted in parameter Z.

Coda lab is great tho, and we have a lot of features in the pipeline that will actually add containerised, reproducible artefacts of your scientific computing work and environment straight from a Jupiter notebook.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact