hex.deeplearning.DeepLearningMojoWriter Maven / Gradle / Ivy

Go to download
Show more of this group Show more artifacts with this name
Show all versions of h2o-algos Show documentation
H2O Algorithms
There is a newer version: 3.46.0.6
package hex.deeplearning;

import hex.ModelMojoWriter;

import java.io.IOException;

import static water.H2O.technote;

public class DeepLearningMojoWriter extends ModelMojoWriter {

  @SuppressWarnings("unused")
  public DeepLearningMojoWriter() {}
  private DeepLearningModel.DeepLearningParameters _parms;
  private DeepLearningModelInfo _model_info;
  private DeepLearningModel.DeepLearningModelOutput _output;

  public DeepLearningMojoWriter(DeepLearningModel model) {
    super(model);
    _parms = model.get_params();
    _model_info = model.model_info();
    _output = model._output;
    if (_model_info.isUnstable()) { // do not generate mojo for unstable model
      throw new UnsupportedOperationException(technote(4, "Refusing to create a MOJO for an unstable model."));
    }
  }

  @Override
  public String mojoVersion() {
    return "1.10";
  }

  @Override
  protected void writeModelData() throws IOException {
    writekv("mini_batch_size", _parms._mini_batch_size);
    writekv("nums", _model_info.data_info._nums);
    writekv("cats", _model_info.data_info._cats);
    writekv("cat_offsets", _model_info.data_info._catOffsets);
    writekv("norm_mul", _model_info.data_info()._normMul);
    writekv("norm_sub", _model_info.data_info()._normSub);
    writekv("norm_resp_mul", _model_info.data_info._normRespMul);
    writekv("norm_resp_sub", _model_info.data_info._normRespSub);
    writekv("use_all_factor_levels", _parms._use_all_factor_levels);
    writekv("activation", _parms._activation);
    writekv("distribution", _parms._distribution);
    boolean imputeMeans=_parms._missing_values_handling.equals(DeepLearningModel.DeepLearningParameters.MissingValuesHandling.MeanImputation);
    writekv("mean_imputation", imputeMeans);
    if (imputeMeans && _model_info.data_info._cats>0) { // only add this if there are categorical columns
      writekv("cat_modes", _model_info.data_info.catNAFill());
    }
    writekv("neural_network_sizes", _model_info.units); // layer 0 is input, last layer is output
    // keep track of neuron network sizes, weights and biases. Layer 0 is the output layer.  Last layer is output layer
    int numberOfWeights = 1+_parms._hidden.length;
    double[] all_drop_out_ratios = new double[numberOfWeights];

    for (int index = 0; index < numberOfWeights; index++) {
      if (index==_parms._hidden.length) { // input layer
        all_drop_out_ratios[index]=0.0;
      } else {
        if (_parms._hidden_dropout_ratios != null) {
          all_drop_out_ratios[index]=_parms._hidden_dropout_ratios[index];
        } else {
          all_drop_out_ratios[index]=0.0;
        }
      }

      //generate hash key to store weights/bias of all layers
      writekv("weight_layer"+index, _model_info.get_weights(index).raw());
      writekv("bias_layer"+index, _model_info.get_biases(index).raw());
    }
    writekv("hidden_dropout_ratios", all_drop_out_ratios);
    writekv("_genmodel_encoding", model.getGenModelEncoding());
    String[] origNames = model._output._origNames;
    if (origNames != null) {
      int nOrigNames = origNames.length;
      writekv("_n_orig_names", nOrigNames);
      writeStringArray(origNames, "_orig_names");
    }
    if (model._output._origDomains != null) {
      int nOrigDomainValues = model._output._origDomains.length;
      writekv("_n_orig_domain_values", nOrigDomainValues);
      for (int i=0; i < nOrigDomainValues; i++) {
        String[] currOrigDomain = model._output._origDomains[i];
        writekv("_m_orig_domain_values_" + i, currOrigDomain == null ? 0 : currOrigDomain.length);
        if (currOrigDomain != null) {
          writeStringArray(currOrigDomain, "_orig_domain_values_" + i);
        }
      }
    }
    writekv("_orig_projection_array", model._output._orig_projection_array);
  }
}