Naively call RedisTF from Python with redis-py

README.md

1. Clone, sh get_deps.sh, and make run https://github.com/lantiga/RedisTF
2. pip install redis tensorflow
3. cd models
4. Run python tf-minimal.py to prepare a minimal graph
5. The attached redistf-py.py demonstrates how to run the test in the README from Python:

$ python redistf-py.py
Setting the graph: OK
Setting tensor t1: OK
Setting tensor t2: OK
Running the thing: OK
Resulting values: ['4', '9']

6. Success!

redistf-py.py

import redis

class RedisTF(object):
    def __init__(self, conn):
        self.__conn = conn

    # Loads a graph
    def SetGraph(self, graph, path):
        with open(path, 'r') as f:
            payload = f.read()
        return self.__conn.execute_command('TF.GRAPH', graph, payload)

    # Sets a tensor
    def SetTensor(self, tensor, dtype, shape, values):
        args = [tensor, dtype]
        args += [str(x) for x in shape]
        args.append('VALUES')
        args += [str(x) for x in values]
        return self.__conn.execute_command('TF.TENSOR', *args)

    # Runs a graph with a list of input tensor-name tuples, storing the result
    def Run(self, graph, inputs, output):
        args = [graph, len(inputs)]
        for i in inputs:
            args.append(i[0])
            args.append(i[1])
        args += [str(x) for x in output]
        return self.__conn.execute_command('TF.RUN', *args)

    # Gets the value of a tensor
    def Values(self, tensor):
        return self.__conn.execute_command('TF.VALUES', tensor)

if __name__ == '__main__':
    conn = redis.StrictRedis()
    rtf = RedisTF(conn)
    print 'Setting the graph: {}'.format(rtf.SetGraph('graph', 'graph.pb'))
    print 'Setting tensor t1: {}'.format(rtf.SetTensor('t1', 'FLOAT', [1, 2], [2, 3]))
    print 'Setting tensor t2: {}'.format(rtf.SetTensor('t2', 'FLOAT', [1, 2], [2, 3]))
    print 'Running the thing: {}'.format(rtf.Run('graph', [('t1', 'a'), ('t2', 'b')], ('t3', 'c')))
    print 'Resulting values: {}'.format(rtf.Values('t3'))