adorationhp / textsum-1

test code for textsum

textsum

• description

  • test code for textsum
  • textsum
    • Text summarization with TensorFlow
    • English Gigaword
    • reference paper A Neural Attention Model for Abstractive Sentence Summarization
      • an implementation using tensorflow, neural-summary-tensorflow

• pre-requesite and setting

  • follow instructions in https://github.com/tensorflow/models/tree/master/textsum

• data format

  • how does data file look like?

  # you need to install google protobuf(http://dchua.com/2016/04/08/installing-grpc,-protobuf-and-its-dependencies-for-python-development/)
  $ python check_data.py --data_path=data/data

  features {
    feature {
      key: "abstract"
      value {
        bytes_list {
          value: "<d> <p> <s> sri lanka closes schools as war escalates . </s> </p> </d>"
        }
      }
    }
    feature {
      key: "article"
      value {
        bytes_list {
          value: "<d> <p> <s> the sri lankan government on wednesday announced the closure of government schools with immediate effect as a military campaign against tamil separatists escalated in the north of the country . </s> <s> t
he cabinet wednesday decided to advance the december holidays by one month because of a threat from the liberation tigers of tamil eelam -lrb- ltte -rrb- against school children , a government official said . </s> <s> `` there are i
ntelligence reports that the tigers may try to kill a lot of children to provoke a backlash against tamils in colombo . </s> <s> `` if that happens , troops will have to be withdrawn from the north to maintain law and order here , \
'\' a police official said . </s> <s> he said education minister richard pathirana visited several government schools wednesday before the closure decision was taken . </s> <s> the government will make alternate arrangements to hold
 end of term examinations , officials said . </s> <s> earlier wednesday , president chandrika kumaratunga said the ltte may step up their attacks in the capital to seek revenge for the ongoing military offensive which she described
as the biggest ever drive to take the tiger town of jaffna . . </s> </p> </d>"
        }
      }
    }
    feature {
      key: "publisher"
      value {
        bytes_list {
          value: "AFP"
        }
      }
    }
  }
  ...

• how to create own training data?

  $ python generate_data.py --input_dir=sample --data_path=sample-0
  $ python check_data.py --data_path=sample-0 --crc=4
  <class 'tensorflow.core.example.example_pb2.Example'>
  features {
    feature {
      key: "abstract"
      value {
        bytes_list {
          value: "<d> <p> <s> hello tensorflow </s> </p> </d>"
        }
      }
    }
    feature {
      key: "article"
      value {
        bytes_list {
          value: "<d> <p> <s> this is a sample file </s> </p> </d>"
        }
      }
    }
  }

  article   \t  <d> <p> <s> this is a sample file </s> </p> </d>
  abstract  \t  <d> <p> <s> hello tensorflow </s> </p> </d>
  ...

• recent version of textsum provides data_convert_example.py

$ python data_convert_example.py --command binary_to_text --in_file data/data --out_file data/text_data
$ python data_convert_example.py --command text_to_binary --in_file data/text_data --out_file data/binary_data
$ python data_convert_example.py --command binary_to_text --in_file data/binary_data --out_file data/text_data2
$ diff data/text_data2 data/text_data

# your text_data format looks like 

abstract=hello world    article=this is a test file
abstract=...            article=....
...

• test

$ ./test.sh -v -v
...
running_avg_loss: 1.002997
running_avg_loss: 1.384698
running_avg_loss: 0.865053
...

• gpu setting tips

# if you want to prevent full gpu-memory allocation, set allow_growth option to 
# `seq2seq_attention.py`, `seq2seq_attention_decode.py`

device_config=tf.ConfigProto(allow_soft_placement=True)
device_config.gpu_options.allow_growth = True
sess = tf.Session(config=device_config)
...