Skip to content

README.md

Latest commit

 

History

History
25 lines (21 loc) · 921 Bytes

README.md

File metadata and controls

25 lines (21 loc) · 921 Bytes

NetReform

NetReform is a tensorflow wrapper for knowledge transfer. It is based on two papers; Net2Net and Network Morphism.

Dependencies

  • tensorflow
  • slim (modified ver.)
  • numpy
  • scipy

Usage

  • See example.py

    from net_reform import NetReform
# model, weights from a previous model
nr = NetReform(model, weights, new_graph)
out_graph, out_session = nr.reform_rand() # NetReform with random values
out_graph, out_session = nr.reform() # NetReform with values derived from net2net or netmorph func.
...
obj = out_graph.get_collection('objective')[0]
out_session.run(obj)

Notes

  • Currently, NetReform only supports Net2Net operators (wider & deeper).