nnet3-init.cc

Go to the documentation of this file.

// nnet3bin/nnet3-init.cc

// Copyright 2012-2015  Johns Hopkins University (author:  Daniel Povey)

// See ../../COPYING for clarification regarding multiple authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//  http://www.apache.org/licenses/LICENSE-2.0
//
// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
// MERCHANTABLITY OR NON-INFRINGEMENT.
// See the Apache 2 License for the specific language governing permissions and
// limitations under the License.

#include "base/kaldi-common.h"
#include "util/common-utils.h"
#include "nnet3/nnet-nnet.h"
#include "hmm/transition-model.h"
#include "tree/context-dep.h"

int main(int argc, char *argv[]) {
  try {
    using namespace kaldi;
    using namespace kaldi::nnet3;
    typedef kaldi::int32 int32;

    const char *usage =
        "Initialize nnet3 neural network from a config file; outputs 'raw' nnet\n"
        "without associated information such as transition model and priors.\n"
        "Search for examples in scripts in /egs/wsj/s5/steps/nnet3/\n"
        "Can also be used to add layers to existing model (provide existing model\n"
        "as 1st arg)\n"
        "\n"
        "Usage:  nnet3-init [options] [<existing-model-in>] <config-in> <raw-nnet-out>\n"
        "e.g.:\n"
        " nnet3-init nnet.config 0.raw\n"
        "or: nnet3-init 1.raw nnet.config 2.raw\n"
        "See also: nnet3-copy, nnet3-info\n";

    bool binary_write = true;
    int32 srand_seed = 0;

    ParseOptions po(usage);
    po.Register("binary", &binary_write, "Write output in binary mode");
    po.Register("srand", &srand_seed, "Seed for random number generator");

    po.Read(argc, argv);
    srand(srand_seed);

    if (po.NumArgs() < 2 || po.NumArgs() > 3) {
      po.PrintUsage();
      exit(1);
    }

    std::string raw_nnet_rxfilename = (po.NumArgs() == 3 ?
                                       po.GetArg(1) : std::string("")),
        config_rxfilename = po.GetArg(po.NumArgs() == 3 ? 2 : 1),
        raw_nnet_wxfilename = po.GetArg(po.NumArgs() == 3 ? 3 : 2);

    Nnet nnet;
    if (po.NumArgs() == 3) {
      ReadKaldiObject(raw_nnet_rxfilename, &nnet);
      KALDI_LOG << "Read raw neural net from "
                << raw_nnet_rxfilename;
    }

    {
      bool binary;
      Input ki(config_rxfilename, &binary);
      KALDI_ASSERT(!binary && "Expect config file to contain text.");
      nnet.ReadConfig(ki.Stream());
    }

    WriteKaldiObject(nnet, raw_nnet_wxfilename, binary_write);
    KALDI_LOG << "Initialized raw neural net and wrote it to "
              << raw_nnet_wxfilename;
    return 0;
  } catch(const std::exception &e) {
    std::cerr << e.what() << '\n';
    return -1;
  }
}


/*
Test script:

cat <<EOF | nnet3-init --binary=false - foo.raw
component name=affine1 type=NaturalGradientAffineComponent input-dim=72 output-dim=59
component name=relu1 type=RectifiedLinearComponent dim=59
component name=final_affine type=NaturalGradientAffineComponent input-dim=59 output-dim=298
component name=logsoftmax type=SoftmaxComponent dim=298
input-node name=input dim=18
component-node name=affine1_node component=affine1 input=Append(Offset(input, -4), Offset(input, -3), Offset(input, -2), Offset(input, 0))
component-node name=nonlin1 component=relu1 input=affine1_node
component-node name=final_affine component=final_affine input=nonlin1
component-node name=output_nonlin component=logsoftmax input=final_affine
output-node name=output input=output_nonlin
EOF

cat <<EOF | nnet3-init --binary=false foo.raw -  bar.raw
component name=affine2 type=NaturalGradientAffineComponent input-dim=59 output-dim=59
component name=relu2 type=RectifiedLinearComponent dim=59
component name=final_affine type=NaturalGradientAffineComponent input-dim=59 output-dim=298
component-node name=affine2 component=affine2 input=nonlin1
component-node name=relu2 component=relu2 input=affine2
component-node name=final_affine component=final_affine input=relu2
EOF

rm foo.raw bar.raw

 */