jubaclient

jubaclient

Jubatus CLI client (unofficial)

Quick Start

# startup jubaclassifier
# For example: 
#  $ docker pull jubatus/jubatus
#  $ docker run -d -p 9199:9199 jubatus/jubatus jubaclassifier -f /opt/jubatus/share/jubatus/example/config/classifier/pa.json

# installation
npm install -g jubaclient

# classifier#train()
echo '[ [ [ "baz", [ [ [ "foo", "bar" ] ] ] ] ] ]' \
| jubaclient classifier train

Requires

• Node.js v6+

Installation

npm install -g jubaclient

Usage

jubaclient service method [-p port] [-h hostname] [-n name] [-t timeoutSeconds]

jubaclient -i [service] [method] [-p port] [-h hostname] [-n name] [-t timeoutSeconds]

jubaclient -v

The jubaclient command requests JSON received from standard input with the specified method to the Jubatus server, and returns the response to the standard output.

JSON passed to standard input is an array of method arguments.

• For methods without arguments it is [].
• If the method argument is a single string type, it should be like [ "foo" ].

Tips: JSON formatting is useful for the jq command.

The command line options are as follows:

• service: sevice name (classifier, nearest_neighbor, etc.)
• method: service method (get_status, train, get_k_center, etc.)
• -p port : port number (default 9199)
• -h hostname : hostname (default localhost)
• -n name : name of target cluster (default '')
• -t timeoutSeconds : timeout (default 0)
• -i : interactive mode
• -v : Print jubaclient's version.

Examples

• #save(id)

  echo '[ "jubaclient_save_1" ]' | jubaclient classifier save 
  

• #get_status()

  echo '[]' | jubaclient classifier get_status | jq '.' 
  

• #get_config()

  echo '[]' | jubaclient classifier get_config | jq '.|fromjson' 
  

• classifier#train(data)

  jubaclient classifier train <<EOF | jq '.'
  [ [ [ "corge", [ [ [ "message", "<p>foo</p>" ] ] ] ] ] ]
  [ [ [ "corge", [ [ [ "message", "<p>bar</p>" ] ] ] ] ] ]
  [ [ [ "corge", [ [ [ "message", "<p>baz</p>" ] ] ] ] ] ]
  [ [ [ "grault", [ [ [ "message", "<p>qux</p>" ] ] ] ] ] ]
  [ [ [ "grault", [ [ [ "message", "<p>quux</p>" ] ] ] ] ] ]
  EOF
  

• classifier#classify(data)

  jubaclient classifier classify <<EOF | jq '.'
  [ [ [ [ [ "message", "<b>quuz</b>" ] ] ] ] ]
  EOF
  

Interactive mode

With the -i option, it will be in interactive mode. When choosing service and method, it provides keyword completion system. When you send Ctrl-C (SIGINT) you return to choosing the service and method, and sending Ctrl-D (EOT) will end the process.

Demonstration

Tutorial

Classifier

See also http://jubat.us/en/tutorial/classifier.html

• start jubaclassifier process.

  jubaclassifier -D --configpath gender.json 
  

• train

  cat train.csv \
  | jq -RcM 'split(",")|[[[.[0],[[["hair",.[1]],["top",.[2]],["bottom",.[3]]],[["height",(.[4]|tonumber)]]]]]]' \
  | jubaclient classifier train
  

• classify

  cat classify.csv \
  | jq -RcM 'split(",")|[[[[["hair",.[0]],["top",.[1]],["bottom",.[2]]],[["height",(.[3]|tonumber)]]]]]' \
  | jubaclient classifier classify \
  | jq '.[]|max_by(.[1])'
  

configure: gender.json

{
  "method": "AROW",
  "converter": {
    "num_filter_types": {}, "num_filter_rules": [],
    "string_filter_types": {}, "string_filter_rules": [],
    "num_types": {}, "num_rules": [],
    "string_types": {
      "unigram": { "method": "ngram", "char_num": "1" }
    },
    "string_rules": [
      { "key": "*", "type": "unigram", "sample_weight": "bin", "global_weight": "bin" }
    ]
  },
  "parameter": { "regularization_weight" : 1.0 }
}

training data: train.csv

male,short,sweater,jeans,1.70
female,long,shirt,skirt,1.56
male,short,jacket,chino,1.65
female,short,T shirt,jeans,1.72
male,long,T shirt,jeans,1.82
female,long,jacket,skirt,1.43

test data: classify.csv

short,T shirt,jeans,1.81
long,shirt,skirt,1.50

Demonstration