For example, if I want to develop a chitchat bot, and I already have a corpus with dialog like the following:
-- Are you a student?
-- No, I am a scientist.
...
Can I train a model directly on this data without going through the regular NLU and Core processes, such as create the NLU data and stories data? In chitchat, it's hard to define the intent, if my corpus is a little large. There can be potentially too many intents. So it seems not to be good to use intent-slot-dm flow to develop the model?
Does RASA provide a way to directly train on the data?
This is currently not possible with the Rasa Stack.
If you model goal oriented use cases, this should be not a big problem. However, for chitchat this can indeed be a bit tricky. Try to see whether you can break down the most frequent chitchat messages in intents, and then handle them in stories.
This may be a question for Survey Monkey, but I felt that someone here may have encountered something like this in past experiences. Is there a way to work with the API of Survey Monkey (SM), to add the information from the survey straight into a database of my own? I realize that I can generate the information into output files, but I was wondering if there was a way to directly access the information from the SM database. I feel like this might cause some privacy concerns for SM. Has anyone attempted this, or would the best option of mine be to create my own surveys without a third party website?
I had a similar issue and here's my solution.
I was doing health related surveys which contain HIPPA protected Personal Health Info. Zapier is NOT HIPAA safe, so the "zap the results over to Google Drive" solution didn't work.
So I wanted a quick n dirty way to grab SM survey data and begin to design a data structure to analyse and store this data. I figured that I would start with <1000 results, sort it out, then build out a bigger/fancier structure as needed.
I just downloaded CSV's of the SM individual responses, munged the downloaded CSV files to make a Python CSV reader happy, then wrote a Python 3.5 script to grab the survey data and spit it out into a couple of output CSV files designed for different analytic purposes.
It was really quick and easy to alter the Python script to deliver different subsets of data to different output files, and really quick and easy to see if these output (CSV or XLS) files really told me what I wanted to know.
This is a really quick and easy way to start analysing right away without spending too much time on procedural overhead. You can alter CSV (or XLS ) tables really quickly and easily, so you can mix and match data / derivative data as much as you want. A wise person once told me "don't think, do." So the more you analyse on small runs of data, the better your final Big Buildout In The Sky will look.
Yah, you can spend a lot of time writing and API and setting up a dbase, but if you are not completely happy with what you want out of the SM data, start small. Hope this helps.
I was assigned an extra credit project where I'm given a set of data and told to run data mining algorithms on them. (Supposed to choose two of Apriori, PART, RIPPER, and J48).
We're told to reuse code for the algorithm implementation of this project, but the only code I can find is example code and can't easily be run on the data set given to us for the project. Does anybody know of premade java data mining algorithms that I can use on my data?
I would start by looking at http://www.cs.waikato.ac.nz/ml/weka/. Websearches suggest that it supports the first two options I checked, Apriori and J48. There is quite reasonable Java source code, but also a GUI interface. You may be able to do what you need just by writing code to produce stuff in a format Weka can read and then working within the Weka GUI.
The application I am working on has "Default lists" one of which is already created in the app currently. The list has events and events touch 2-3 other models. Which would make seeding, etc very time consuming due to the complexity of the lists and the associated models the list has data in
Due to the complexity of the lists I would prefer to build the lists though the UI and then extracting it for later use.
Is there any worthwhile way of extracting the aforementioned list object and for lack of a better term "bootstrap it"
Thanks for your help in advance.
I think what you are trying to get at is seed data. Take a look at this railscats on just that.
Solution: https://github.com/rhalff/seed_dump
I highly enjoy the comments on the github page
It mainly exists for people who are too lazy writing create statements
in db/seeds.rb themselves and need something (seed_dump) to dump data
from the table(s) into seeds.rb
My response to that is "work smart, not hard" no need for me to spend a day or 2 writing out long seeds instead of doing actual work.
Unless i'm hungover then i'll just pretend seed_dump is on the fritz ;)
I've read a statement somewhere that generating UI automatically from DB layout (or business objects, or whatever other business layer) is a bad idea. I can also imagine a few good challenges that one would have to face in order to make something like this.
However I have not seen (nor could find) any examples of people attempting it. Thus I'm wondering - is it really that bad? It's definately not easy, but can it be done with any measure success? What are the major obstacles? It would be great to see some examples of successes and failures.
To clarify - with "generating UI automatically" I mean that the all forms with all their controls are generated completely automatically (at runtime or compile time), based perhaps on some hints in metadata on how the data should be represented. This is in contrast to designing forms by hand (as most people do).
Added: Found this somewhat related question
Added 2: OK, it seems that one way this can get pretty fair results is if enough presentation-related metadata is available. For this approach, how much would be "enough", and would it be any less work than designing the form manually? Does it also provide greater flexibility for future changes?
We had a project which would generate the database tables/stored proc as well as the UI from business classes. It was done in .NET and we used a lot of Custom Attributes on the classes and properties to make it behave how we wanted it to. It worked great though and if you manage to follow your design you can create customizations of your software really easily. We also did have a way of putting in "custom" user controls for some very exceptional cases.
All in all it worked out well for us. Unfortunately it is a sold banking product and there is no available source.
it's ok for something tiny where all you need is a utilitarian method to get the data in.
for anything resembling a real application though, it's a terrible idea. what makes for a good UI is the humanisation factor, the bits you tweak to ensure that this machine reacts well to a person's touch.
you just can't get that when your interface is generated mechanically.... well maybe with something approaching AI. :)
edit - to clarify: UI generated from code/db is fine as a starting point, it's just a rubbish end point.
hey this is not difficult to achieve at all and its not a bad idea at all. it all depends on your project needs. a lot of software products (mind you not projects but products) depend upon this model - so they dont have to rewrite their code / ui logic for different client needs. clients can customize their ui the way they want to using a designer form in the admin system
i have used xml for preserving meta data for this sort of stuff. some of the attributes which i saved for every field were:
friendlyname (label caption)
haspredefinedvalues (yes for drop
down list / multi check box list)
multiselect (if yes then check box
list, if no then drop down list)
datatype
maxlength
required
minvalue
maxvalue
regularexpression
enabled (to show or not to show)
sortkey (order on the web form)
regarding positioning - i did not care much and simply generate table tr td tags 1 below the other - however if you want to implement this as well, you can have 1 more attribute called CssClass where you can define ui specific properties (look and feel, positioning, etc) here
UPDATE: also note a lot of ecommerce products follow this kind of dynamic ui when you want to enter product information - as their clients can be selling everything under the sun from furniture to sex toys ;-) so instead of rewriting their code for every different industry they simply let their clients enter meta data for product attributes via an admin form :-)
i would also recommend you to look at Entity-attribute-value model - it has its own pros and cons but i feel it can be used quite well with your requirements.
In my Opinion there some things you should think about:
Does the customer need a function to customize his UI?
Are there a lot of different attributes or elements?
Is the effort of creating such an "rendering engine" worth it?
Okay, i think that its pretty obvious why you should think about these. It really depends on your project if that kind of model makes sense...
If you want to create some a lot of forms that can be customized at runtime then this model could be pretty uselful. Also, if you need to do a lot of smaller tools and you use this as some kind of "engine" then this effort could be worth it because you can save a lot of time.
With that kind of "rendering engine" you could automatically add error reportings, check the values or add other things that are always build up with the same pattern. But if you have too many of this things, elements or attributes then the performance can go down rapidly.
Another things that becomes interesting in bigger projects is, that changes that have to occur in each form just have to be made in the engine, not in each form. This could save A LOT of time if there is a bug in the finished application.
In our company we use a similar model for an interface generator between cash-software (right now i cant remember the right word for it...) and our application, just that it doesnt create an UI, but an output file for one of the applications.
We use XML to define the structure and how the values need to be converted and so on..
I would say that in most cases the data is not suitable for UI generation. That's why you almost always put a a layer of logic in between to interpret the DB information to the user. Another thing is that when you generate the UI from DB you will end up displaying the inner workings of the system, something that you normally don't want to do.
But it depends on where the DB came from. If it was created to exactly reflect what the users goals of the system is. If the users mental model of what the application should help them with is stored in the DB. Then it might just work. But then you have to start at the users end. If not I suggest you don't go that way.
Can you look on your problem from application architecture perspective? I see you as another database terrorist – trying to solve all by writing stored procedures. Why having UI at all? Try do it in DB script. In effect of such approach – on what composite system you will end up? When system serves different businesses – try modularization, selectively discovered components, restrict sharing references. UI shall be replaceable, independent from business layer. When storing so much data in DB – there is hard dependency of UI – system becomes monolith. How you implement MVVM pattern in scenario when UI is generated? Designers like Blend are containing lots of features, which cannot be replaced by most futuristic UI generator – unless – your development platform is Notepad only.
There is a hybrid approach where forms and all are described in a database to ensure consistency server side, which is then compiled to ensure efficiency client side on deploy.
A real-life example is the enterprise software MS Dynamics AX.
It has a 'Data' database and a 'Model' database.
The 'Model' stores forms, classes, jobs and every artefact the application needs to run.
Deploying the new software structure used to be to dump the model database and initiate a CIL compile (CIL for common intermediate language, something used by Microsoft in .net)
This way is suitable for enterprise-wide software and can handle large customizations. But keep in mind that this approach sets a framework that should be well understood by whoever gonna maintain and customize the application later.
I did this (in PHP / MySQL) to automatically generate sections of a CMS that I was building for a client. It worked OK my main problem was that the code that generates the forms became very opaque and difficult to understand therefore difficult to reuse and modify so I did not reuse it.
Note that the tables followed strict conventions such as naming, etc. which made it possible for the UI to expect particular columns and infer information about the naming of the columns and tables. There is a need for meta information to help the UI display the data.
Generally it can work however the thing is if your UI just mirrors the database then maybe there is lots of room to improve. A good UI should do much more than mirror a database, it should be built around human interaction patterns and preferences, not around the database structure.
So basically if you want to be cheap and do a quick-and-dirty interface which mirrors your DB then go for it. The main challenge would be to find good quality code that can do this or write it yourself.
From my perspective, it was always a problem to change edit forms when a very simple change was needed in a table structure.
I always had the feeling we have to spend too much time on rewriting the CRUD forms instead of developing the useful stuff, like processing / reporting / analyzing data, giving alerts for decisions etc...
For this reason, I made long time ago a code generator. So, it become easier to re-generate the forms with a simple restriction: to keep the CSS classes names. Simply like this!
UI was always based on a very "standard" code, controlled by a custom CSS.
Whenever I needed to change database structure, so update an edit form, I had to re-generate the code and redeploy.
One disadvantage I noticed was about the changes (customizations, improvements etc.) done on the previous generated code, which are lost when you re-generate it.
But anyway, the advantage of having a lot of work done by the code-generator was great!
I initially did it for the 2000s Microsoft ASP (Active Server Pages) & Microsoft SQL Server... so, when that technology was replaced by .NET, my code-generator become obsoleted.
I made something similar for PHP but I never finished it...
Anyway, from small experiments I found that generating code ON THE FLY can be way more helpful (and this approach does not exclude the SAVED generated code): no worries about changing database etc.
So, the next step was to create something that I am very proud to show here, and I think it is one nice resolution for the issue raised in this thread.
I would start with applicable use cases: https://data-seed.tech/usecases.php.
I worked to add details on how to use, but if something is still missing please let me know here!
You can change database structure, and with no line of code you can start edit data, and more like this, you have available an API for CRUD operations.
I am still a fan of the "code-generator" approach, and I think it is just a flavor of using XML/XSLT that I used for DATA-SEED. I plan to add code-generator functionalities.