Skip to content

Latest commit

 

History

History
111 lines (67 loc) · 2.44 KB

README.rst

File metadata and controls

111 lines (67 loc) · 2.44 KB
Raw

jubadump

jubadump converts saved jubatus internal model to json. jubadump supports only classifier (except for NN), regression, recommender (inverted_index), anomaly (lof + inverted_index) and nearest_neighbor model.

Requirements

  • g++
  • pkg-config
  • jubatus_core
  • msgpack

Compile and Install

$ ./waf configure
$ ./waf
$ ./waf install

Usage

  1. Call save RPC and save your model.
  2. Execute jubadump.
usage: jubadump --input=string [options] ...
options:
  -i, --input    Input file (string)
  -?, --help     print this message

Format

  • common
    • weights (some values are available only when IDF/BM25 weighting is used)
      • version_number : Version of model. This value will be updated by MIX.
      • document_frequencies : Frequency of each feature in data inputted so far.
      • document_count : Number of all documents. This value will be used in calculation of global_weight(idf).
      • group_frequencies : Frequency of each Datum (string_values) key in data inputted so far.
      • group_total_length : Number of total features extracted from each Datum (string_values) key in data inputted so far.
  • classifier
    • storage
      • labels
        • version_number : Version of model. This value will be updated by MIX.
        • label_count : Number of traind data of each label.
      • storage
        • weight : Weights of each feature and each label in data inputted so far.
  • regression
    • storage
      • weight : Weights of each feature and each label in data inputted so far.
  • recommender (inverted_index)
    • index
      • storage
        • inv: No data
      • original
        • inv : Features of each ID
  • anomaly (lof based on inverted_index)
    • storage
      • lof_records
        • entry: kdist and lrd value of each ID
      • nn_records : Refer to recommender for the list of values.
  • nearest_neighbor
    • table
      • clock: generation of the table (number of updates)
      • schema: schema of the table
      • data: records of each ID
        • version: generation of the record
        • values: data of each column for the ID; multiple 64-bit integers are used to represent the bit vector for column longer than 64 bit.
        • owner: identifier of the server who has a permission to modify the record

License

LGPL 2.1