AI, Machine Learning, Neural Networks, TensorFlow
Hello,
I would like to start interacting with ML to get ahead of the competition, and I'm curious, is there a way to implement TensorFlow or something similar to the Mango as a module? Is this line of development visualized for the future?
Last week, I was exploring Github repositories related to Infinite Automation, and found that @terrypacker has worked in an experimental ML module. Am I right? Is the project still in development?
Thanks!
I have in the past played around with a few different ways to bring ML into Mango and found that our scripting infrastructure is probably going to be the best path forward. The examples you have seen were my attempts to do some modeling directly in Java which does work nicely but is quite time consuming since many features that are built into ML languages have to be hand coded in Java. I've considered adding support for Python and R but we haven't really enough interest yet to drive the development.
I took a look a t TensorFlow and it has a javascript implementation tensorflow.js which could possibly be used in our java scripting environment. However I tried to download and add the minified version as a global script and it didn't work so I will defer to @Jared-Wiltshire to comment on this and then he and I will discuss how this could potentially be done.
@terrypacker said in AI, Machine Learning, Neural Networks, TensorFlow:
I took a look a t TensorFlow and it has a javascript implementation tensorflow.js which could possibly be used in our java scripting environment. However I tried to download and add the minified version as a global script and it didn't work so I will defer to @Jared-Wiltshire to comment on this and then he and I will discuss how this could potentially be done.
It is highly likely that it wont work in Nashorn (the Java JS engine) as it doesn't support the latest ECMAScript features. Not to mention that there may be memory/script size limits.
Hello, @terrypacker and @Jared-Wiltshire.
@terrypacker said in AI, Machine Learning, Neural Networks, TensorFlow:
The examples you have seen were my attempts to do some modeling directly in Java which does work nicely but is quite time consuming since many features that are built into ML languages have to be hand coded in Java. I've considered adding support for Python and R but we haven't really enough interest yet to drive the development.
Interesting, it's pretty cool that you've tried it. Is it possible that something cloud-based, maybe a combo similar to lambda functions + Amazon API Gateway, could make the ML model more available and reduce the need for Python or R support in the same environment?
I have no idea if this could be done, but if the answer is yes, I think it could be charged as an extra service/module. Still, I would totally understand if the development doesn't start until more people show interest, which I hope happens soon.
@terrypacker said in AI, Machine Learning, Neural Networks, TensorFlow:
However I tried to download and add the minified version as a global script and it didn't work.
I was thinking about trying that, thanks for saving me the test time.
@jared-wiltshire said in AI, Machine Learning, Neural Networks, TensorFlow:
It is highly likely that it wont work in Nashorn (the Java JS engine) as it doesn't support the latest ECMAScript features. Not to mention that there may be memory/script size limits.
Hmmm. I understand. Well, maybe the memory/size limits could also be avoided with the cloud-based approach, right?
Alright, thanks a lot for your time and answers, guys!
@jflores13 said in AI, Machine Learning, Neural Networks, TensorFlow:
Hmmm. I understand. Well, maybe the memory/size limits could also be avoided with the cloud-based approach, right?
Yes, no problem running TensorFlow JS in the cloud on NodeJS. However I am not sure what advantage this would afford over the native binaries.
@jflores13 said in AI, Machine Learning, Neural Networks, TensorFlow:
Interesting, it's pretty cool that you've tried it. Is it possible that something cloud-based, maybe a combo similar to lambda functions + Amazon API Gateway, could make the ML model more available and reduce the need for Python or R support in the same environment?
Something to look into in the future for sure, but that approach would have its own problems to overcome. For example getting the data from Mango to the model.
@jared-wiltshire said in AI, Machine Learning, Neural Networks, TensorFlow:
Yes, no problem running TensorFlow JS in the cloud on NodeJS. However I am not sure what advantage this would afford over the native binaries.
True, I'm not sure either, just trying to think about all the possibilities, hehe.
@terrypacker said in AI, Machine Learning, Neural Networks, TensorFlow:
@jflores13 said in AI, Machine Learning, Neural Networks, TensorFlow:
Something to look into in the future for sure, but that approach would have its own problems to overcome. For example getting the data from Mango to the model.
Yeeeah, I don't visualize how that interaction could be simplified. I guess it's just not as simple as I was hoping.
I'll keep looking for possible solutions to this and will share anything I can find.
Thanks!
