I posted this earlier but all the traffic crashed my basic test server - Im new to HN and under estimated the traffic. So I set it up on an powerful ec2 image - lets see if this can handle it. Since Im paying for the ec2 hosting by the hour, I will on leave this link up for a few days. Thank you for your feedback.
Original message:
This is a VERY EARLY demo of a web application I am building. Im trying to put statistical software on the web (almost all stat software is desktop). The first page is a document management module (similar to the one on Google Docs)-you can drag items into folders, share folders with other users, trash them, etc. Double clicking on a data set takes you to the data analysis module, where you can do statistical analysis. Right now I have only a simple bivariate regression module up - I want to keep it simple for the demo/proof of concept.
I dont expect to be accepted by YC but still want some feed back from the YC community. Please dont be kind - tell me everything that is wrong - but be constructive!
I did manage to see it before it disappeared first time around.
It mostly worked nicely for me but crashed somewhere in datamanagement.nocache.js
As a proof of concept demo is ok. However, it would be better to have a demo with the full interface even if some parts are not working so that you can demonstrate the full workflow. (Upload file - manipulate data for example throw away a row, choose statistic, plot save, print). Looks like is build using EXTJS are you using R for calcs?
First, a meta-note -- if you want app reviews, don't mention that they're YC app submissions as that usually gets them killed. (Since otherwise there would be a bajillion apps getting submitted here.)
In general it's slick, presumably a good chunk of that coming from GWT usage. However, I kind of lost interest after a couple minutes flipping through stuff. There's no pain point that jumps out at me, i.e., "Oh, wow, I'd love to use that for [...]!"
It's a natural tendency for geeks (myself included) to build platforms to solve problems rather than solving problems directly. Fight that. :-)
There are some good posts on UI on codinghorror.com, but the site is not dedicated to it -- you'd have to dive in to find them.
Also, try to look at how others do their web applications. 37 Signals' stuff may be inspirational, and you can look at how Gmail and Google Reader work.
I don't know if that's the right approach. If his target audience are people who don't know the difference between Web 2.0 and 1.0 (say, accountants in the accounting department of a big corporation), they would probably feel more at home with the desktop metaphors.
Case in point, Google wants people to use Google Docs like their Office suite, except it's 100% web based. But to ease the transition they have this notion of "folders" in G Docs even know it could very well just be a bunch of tags like in Gmail. When non Web savvy people see "folders" they go "ah, I get it."
I didn't think of that. You're right -- knowing your users' background and knowledge is very important.
However, there are things that I believe simply shouldn't be carried over from desktop to web.
Even in desktop, modal dialogs reek of "I don't know or care how to do it better". Double clicks have a similar feel to me. And neither is much present on the web (1.0 and 2.0).
Even GDocs don't rely on double clicking (though they try to mimic the Office behaviour in other ways).
But sure, if the users never really used the web and are used to desktop tools, it would be easier for them if this program mimicked that.
Neat idea. But I wonder, what are you offering that excel doesn't? So far, the only tool I can see is
y vs x and thats not tough to do in excel. I'm not saying you aren't offering anything, but I don't know what it is.
For the interface, a few things I didn't like, but others can chime in.
1) When I click on "Basic Statistics" I see the tests there. Then, when I click on "Multivariate Statistics" the basic title bar drops down to the bottom. If I were trying to look for a test to use, I would want to scroll through each category. They way you have it makes that tough.
2) When I go to analyze a data set, how do I get back to the list of files? I figured it out (click on Data Management) but that should be easier to find.
Also, something that I think would be very cool would be to allow me to take data sets from different files. This would allow me to use your app as a storage for my data sets. I could organize files however made sense to me and then arbitrarily go there and look for correlations. "Oh, I wonder if there is a correlation between temperature and mortality. Lets take the weather file and the mortality file and check."
"Neat idea. But I wonder, what are you offering that excel doesn't?"
Right now nothing, but excel doesnt do (easily) the more advanced statistical procedures that will be added - for example fitting reliability data to a distribution of failure times, etc.
Thanks for pointing out (1). I will have think about this setup.
You also make a good point above mixing access to different data sets. At this point, its almost as if your asking if I will implement the ability for users to connect to a database. Yes in time.
By the way Im open to meeting another cofounder. My current cofounder who used to work at SAS can only commit maybe 10-20% of his time (he's working on a PhD in stat). If you are interested in statistical software please let me know below.
As a general tip, any time you feel the need to show a message telling the user how to interact with your application ("double click on X to ...") you should reconsider. Besides which, double clicking isn't a good UI mechanism for web apps (or any other apps, really).
Thanks. I agree, I felt uneasy about this whole double click thing. It will be gone in the next update. Do you recommed a special button like "launch" instead?
What does this app do? I clicked "start", and was plopped down on the home page with a frightening amount of information. It would be nice if you explained in a sentence or two what the app does, and why I should use it.
I see that while I clicked away to write this I got the "welcome to statgadgets" popup. I still don't know what the data sets are or why I would want to "analyze" them.
This is really nice. Took me a moment to figure out how to use (eg dragging the fields for the graphs), but nothing too tricky. Over all, its nice and looks like it could end up being quite useful, going by the other (as yet unimplemented) menus). Keep it up!
As a heads-up, let me tell you that I'm applying to YC with a similar concept under the name Theoryville. We envision a very different interface, but we haven't built a demo yet. I'd be happy to exchange ideas with you privately; I'm not difficult to track down.
Here are some specific criticisms: I don't like the interface's reliance on double-clicking. I don't like that datasets open in a new tab. I expect words that are bright colors to be clickable (for instance, the word "data" under each dataset's name). When I "open" a dataset, I don't expect to be taken directly to the analysis screen; I expect to learn more about the dataset. For the output to be useful, I really need to be able to save the table in some way. The linear fit line should be shown by default. Also, there seems to be no way to remove the fit line once you've added it.
All in all, I agree with Tomas's view that it should be more "webby." And that's a non-trivial problem to solve for something as complex as statistical analysis. Best of luck!
(Also, have you seen Swivel? They don't do much in the way of data analysis, but they have a great interface for doing data visualizations. Definitely check them out.)
27 comments
[ 2.7 ms ] story [ 71.4 ms ] threadOriginal message:
This is a VERY EARLY demo of a web application I am building. Im trying to put statistical software on the web (almost all stat software is desktop). The first page is a document management module (similar to the one on Google Docs)-you can drag items into folders, share folders with other users, trash them, etc. Double clicking on a data set takes you to the data analysis module, where you can do statistical analysis. Right now I have only a simple bivariate regression module up - I want to keep it simple for the demo/proof of concept.
I dont expect to be accepted by YC but still want some feed back from the YC community. Please dont be kind - tell me everything that is wrong - but be constructive!
Might not matter for YC demo; might matter lots to prove awareness of landing pages or whatever.
I'm a total number retard so I'm afraid I didn't last long in the application.
It mostly worked nicely for me but crashed somewhere in datamanagement.nocache.js
As a proof of concept demo is ok. However, it would be better to have a demo with the full interface even if some parts are not working so that you can demonstrate the full workflow. (Upload file - manipulate data for example throw away a row, choose statistic, plot save, print). Looks like is build using EXTJS are you using R for calcs?
yes and yes. Also Im also using some python along with R (im also using a R-python interface). Django is the web framework.
In general it's slick, presumably a good chunk of that coming from GWT usage. However, I kind of lost interest after a couple minutes flipping through stuff. There's no pain point that jumps out at me, i.e., "Oh, wow, I'd love to use that for [...]!"
It's a natural tendency for geeks (myself included) to build platforms to solve problems rather than solving problems directly. Fight that. :-)
Double clicks, folders, dialog windows, etc. go against the web experience. Try to make the user interface more "webby" -- it will benefit everyone.
I think it's a great source to get you started.
There are some good posts on UI on codinghorror.com, but the site is not dedicated to it -- you'd have to dive in to find them.
Also, try to look at how others do their web applications. 37 Signals' stuff may be inspirational, and you can look at how Gmail and Google Reader work.
FogBugz can also nudge you into the right direction: http://www.fogcreek.com/FogBugz/
Case in point, Google wants people to use Google Docs like their Office suite, except it's 100% web based. But to ease the transition they have this notion of "folders" in G Docs even know it could very well just be a bunch of tags like in Gmail. When non Web savvy people see "folders" they go "ah, I get it."
imho.
However, there are things that I believe simply shouldn't be carried over from desktop to web.
Even in desktop, modal dialogs reek of "I don't know or care how to do it better". Double clicks have a similar feel to me. And neither is much present on the web (1.0 and 2.0).
Even GDocs don't rely on double clicking (though they try to mimic the Office behaviour in other ways).
But sure, if the users never really used the web and are used to desktop tools, it would be easier for them if this program mimicked that.
For the interface, a few things I didn't like, but others can chime in.
1) When I click on "Basic Statistics" I see the tests there. Then, when I click on "Multivariate Statistics" the basic title bar drops down to the bottom. If I were trying to look for a test to use, I would want to scroll through each category. They way you have it makes that tough.
2) When I go to analyze a data set, how do I get back to the list of files? I figured it out (click on Data Management) but that should be easier to find.
Also, something that I think would be very cool would be to allow me to take data sets from different files. This would allow me to use your app as a storage for my data sets. I could organize files however made sense to me and then arbitrarily go there and look for correlations. "Oh, I wonder if there is a correlation between temperature and mortality. Lets take the weather file and the mortality file and check."
"Neat idea. But I wonder, what are you offering that excel doesn't?"
Right now nothing, but excel doesnt do (easily) the more advanced statistical procedures that will be added - for example fitting reliability data to a distribution of failure times, etc.
Thanks for pointing out (1). I will have think about this setup.
You also make a good point above mixing access to different data sets. At this point, its almost as if your asking if I will implement the ability for users to connect to a database. Yes in time.
I see that while I clicked away to write this I got the "welcome to statgadgets" popup. I still don't know what the data sets are or why I would want to "analyze" them.
As a heads-up, let me tell you that I'm applying to YC with a similar concept under the name Theoryville. We envision a very different interface, but we haven't built a demo yet. I'd be happy to exchange ideas with you privately; I'm not difficult to track down.
Here are some specific criticisms: I don't like the interface's reliance on double-clicking. I don't like that datasets open in a new tab. I expect words that are bright colors to be clickable (for instance, the word "data" under each dataset's name). When I "open" a dataset, I don't expect to be taken directly to the analysis screen; I expect to learn more about the dataset. For the output to be useful, I really need to be able to save the table in some way. The linear fit line should be shown by default. Also, there seems to be no way to remove the fit line once you've added it.
All in all, I agree with Tomas's view that it should be more "webby." And that's a non-trivial problem to solve for something as complex as statistical analysis. Best of luck!
(Also, have you seen Swivel? They don't do much in the way of data analysis, but they have a great interface for doing data visualizations. Definitely check them out.)