qlai / EasyTensor

Project for HackCambridge 2017 - GUI for basic models in Tensorflow

###HackCambridge 2017 Project by Dao Zhou, Minxuan Xie, Renqiao Zhang, Qiuying Lai ##EasyTensor: GUI for Simple Models in Tensorflow Helping beginners and students to understand the structure of Tensorflow better by providing a GUI where standard neural net models written in Tensorflow (Python) can be generated.

This uses MNIST Data as an example. debug*.py shows how the models can be trained and used.

Dependencies: tensorflow, jinja2, flask, opencv

Examples:

• To draw a model use the following on terminal: cd server python easy_tensor_server.py to launch drawer on localhost
• Click generate on the webpage to generate model file out local directory, this will output a file in 'models' directory in the format [model_type]_output.py
• Follow template on debug0 for a MultiLayerPerceptron Model, debug1 to run a CNN Model
• Launch Tensorboard using tensorboard --logdir='[your dir]' to visualize in TF graph to check correctness
• use models/predict.py to predict new digit images

Further Instructions:

• You must use a perceptron model after layers of CNN for flattening before output layer