RNN Learner (Text)
This app shows how to implement a simple Recurrent Neural Network (RNN). Here is how you can train this app to learn from a given text file and try to predict some output:
1. Click on Add New and select a text file.
2. It will show a pop up where you can modify the number of hidden layers you want for your training set.
3. Also you can select whether you want to build the vocabulary based on characters or words. (by characters: {A, B, C, 1, 2, 3,...} and by words: {Hello, World, How are, you, ...}
4. Simple add your training set and start training.
5. Click on "Predict" button to generate some output based on what it has learned so far.
If you want to know the technical functionality of the RNN or how to implement RNN, you can click on "Knowledge Base" in the app.
Suggestion: I know that mobile phone is not a good platform to get the best out of RNN. But you can get really good output quickly if you create the training sets using small text files. How fast your phone will train itself, depends on the processing speed of your phone.
Category : Education
Reviews (9)
Nice App! The trains-counter is a bit buggy, and a new feature I'd like was a graph of the loss or sth else.
It needs .txt files, which I don't have. I HATE IT! IY IZ SL BAED!!!!!!!
Alright, i gave this app another go, it didn't work again. I'm still having my 24hr+ generations fail. They keep having the same end prediction behaviors. Here's some predictions from my one on Titanium Dioxide. ANDTITANIUMTITANIUMTITANIUMTITANIUM COLORINGTITANIUMTITANIUMTITANIUMTITANIUM FOURTEENTITANIUMTITANIUMTITANIUMTITANIUM PIGMENTSTITANIUMTITANIUMTITANIUMTITANIUM
Neat, if niche, app idea and maybe the only RNN predictive text generator that's easy to get running - I've seen some Windows programs for this purpose on GitHub and they've got a ton of confusing prerequisite things and prerequisites for the prerequisites, that need to be installed first. Here a RNN is all in one little easy-to-install app [although wish it were possible for me to run it on a faster device than what I'm using!] It'd be good to have a little more control over settings, though - and a clearer estimate quickly, early in the process, of how long it'll take to get an 'optimal' or near-optimal result. Your documentation also covers a lot of educational ground relating to neural networks but doesn't clearly or directly clarify all the settings of the app itself. Nonetheless, thanks for making this - especially given that it's freeware. That's pretty awesome.
It's cool but when I ask it to predict it only says one word, am I doing something wrong, it's been training for almost 13 hours. Thanks for the immediate response, and yes I'm new to RNNS :)
Cool. But sample text with detailed into would help a lot. I mean, actual text, intervals triained etc... Not enough info
You should update this to use pdfs pictures audio and video
It is a nice RNN learner.
very usefull, but i'm having a problem, that being that it's still not doing much after having trained for probably around 35 hours, it won't show how much time it's spent anymore, and its efficiency isnt changing. am i just not training it long enough? the info about RNNs is usefull, but i wish there would be in entry to tell you about the interface without having to explain RNNs in detail, e.g. what does "loss" mean, what should i expect as i train it, and generally how to use the app