Home
This is a python visual programming tool based on tensorflow. It's in the early stages. I wrote it over Christmas 2017. The idea is to wrap tensorflow in a visual programming GUI so that you can play around with it more easily. The code is a bit messy so I'll have to refactor it at some point. Its all in python and I've tested it on my MacBook Air but should run on anything. It takes a tensor and tries to guess the best representation. For example a tensor with shape (200,200,3) is most likely an RGB image. The ultimate goal is to make it so you can visually build neural networks and train them.
You will need to install tensorflow and some other libraries. It uses python2.
To run it use
python2 start.py
To create constants or variables type the value in the input box. Examples are:
3.2423
[1,2],[3,4]
You can also put in python expressions. So to create a 100x100 matrix, try one of these:
np.zeros([100,100])
np.full([100,100],0.2)
You can also select the type such as int or float32 from the drop-down. Then click on constant, variable or sometimes if you need a non-tensorflow value list.
To create new nodes, you can select one of the common ones like add or multiply or choose one from the drop down lists.
There is no flow control yet. It simply updates all the nodes every frame by getting there values. Starting with the nodes at the bottom of the graph.
A mandelbrott fractal example:
To get the variable node in this example type np.zeros([128,128]) into the input box, select complex64 from the type drop-down then click variable. (This needs to be made simpler at some point!)
- Loading and saving