hulalazz / chainer-graphviewer

Graph viewer on Jupyter Notebook for Chainer

Install

$ pip install git+https://github.com/disktnk/chainer-graphviewer

• This tool is run on Jupyter Notebook
• if view ONNX model, need $ pip install onnx
• if view Chainer model, need $ pip install chainer
• if only use show_graph, Tensorflow user for example, don't have to install ONNX or Chainer module

Example for Chainer

# Network definition
class Net(chainer.Chain):
    def __init__(self):
        pass
    # snip...

from graphviewer.notebook.show import show_graph
from graphviewer.parser.chainer_graph import get_graphdef_from_model

model = L.Classifier(Net())

x = chainer.Variable(numpy.random.rand(1, 1, 28, 28).astype(numpy.float32))
t = chainer.Variable(numpy.random.rand(1).astype(numpy.int32))
y = model(x, t)

gdef = get_graphdef_from_model(model, (x, t))
show_graph(gdef)

Example for ONNX

show model.onnx

from graphviewer.notebook.show import show_graph
from graphviewer.parser.onnx_graph import get_graphdef_from_file

gdef = get_graphdef_from_file('model.onnx')
show_graph(gdef)

ipynb Example

• example_for_module
  • running example after install this repository
• example_for_code_explain
  • running example with detail scripts

Prepare (for developer)

1. copy *.proto, current version is v1.12.0, put graphviewer/proto/
2. $sed -i -e 's|"tensorflow/core/framework|"graphviewer/proto|g' graphviewer/proto/*.proto
3. protoc --python_out=./ ./graphviewer/proto/*.proto