Neural Network Management Tool applied to 1 Neural Network
$750-1500 USD
Paid on delivery
I need to have a reusable environment done in python to train several neural networks with train and validation data provided from a Database and for storing neural network configurations/weights and associated performance data in same DB.
The goal is to develop also a (visual) interface from which I can see if the underlying code has matching interfaces. It is thereby assumed that the used code responsible for providing the data to the input, extracting it from the output, applying a fitness function on the output, defining the Neural Network structure and/or storing the weights of the connections has some comments in it that is then documenting the relevant interfaces.
Additionally it is assumes that all information relevant to run a Neural Network with the training/validation data and outputting these data, are stored in the DB as well. Each run should be considered as an “experiment” which is being documented via data within the DB.
I have already an Architecture and a Database model that I would share with qualified developers. Most of the Data-Layer is already being done using Python 3.6.
The entire project should be done in Python. I don’t care about visually pleasing interfaces, I am much more interested about functionality that delivers a tool that allows me to get Neural Networks defined while its training and validation output can relatively easily being examined.
Therefore, I prefer someone who has already experience with NN’s in particular has experiences on how to define neural network/deep learning architectures in python and store data in a DB and know what is required to do when evaluating the quality of NN's.
I want to start with one neural network framework (i.e. TensorFlow or Theano or Keras) applied to 1 concrete NN for which I have training/validation data. As part of this project I need to have this NN implemented in a way so that the data are being stored in the DB. The neural network training should be done using back-propagation.
Input values will/should be stored in a compressed array turned into a string (compress + binary to text transformation) and stored within a varchar column of a table). Same should be done with neural network architecture data and the output values and the connection weights.
The system to be developed should have functions that store and restore weights from the NN to a DB and from DB back to the neural network.
Focus should also be on providing an relatively easy & efficient (reusable/template kind of) code that helps me to manage, input, output and multiple hidden layer and the connection between them … I expect that numpy, panda etc. is being used.
Project ID: #14628241
