geco_data_generator / examples /generate-data-japanese.py
bstds's picture
Upload 53 files
16dcd38
raw
history blame contribute delete
No virus
6.08 kB
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
from geco_data_generator import basefunctions, attrgenfunct, contdepfunct, generator, corruptor
import random
random.seed(42)
# Set the Unicode encoding for this data generation project. This needs to be
# changed to another encoding for different Unicode character sets.
# Valid encoding strings are listed here:
# http://docs.python.org/library/codecs.html#standard-encodings
#
unicode_encoding_used = 'cp932'
# The name of the record identifier attribute (unique value for each record).
# This name cannot be given as name to any other attribute that is generated.
#
rec_id_attr_name = 'rec-id'
# Set the file name of the data set to be generated (this will be a comma
# separated values, CSV, file).
#
out_file_name = 'example-data-japanese.csv'
# Set how many original and how many duplicate records are to be generated.
#
num_org_rec = 10000
num_dup_rec = 10000
# Set the maximum number of duplicate records can be generated per original
# record.
#
max_duplicate_per_record = 3
# Set the probability distribution used to create the duplicate records for one
# original record (possible values are: 'uniform', 'poisson', 'zipf').
#
num_duplicates_distribution = 'zipf'
# Set the maximum number of modification that can be applied to a single
# attribute (field).
#
max_modification_per_attr = 1
# Set the number of modification that are to be applied to a record.
#
num_modification_per_record = 5
# Check if the given the unicode encoding selected is valid.
#
basefunctions.check_unicode_encoding_exists(unicode_encoding_used)
# -----------------------------------------------------------------------------
# Define the attributes to be generated (using methods from the generator.py
# module).
#
surname_attr = generator.GenerateFreqAttribute(
attribute_name='surname',
freq_file_name='surname-freq-japanese.csv',
has_header_line=False,
unicode_encoding=unicode_encoding_used,
)
credit_card_attr = generator.GenerateFuncAttribute(
attribute_name='credit-card-number', function=attrgenfunct.generate_credit_card_number
)
age_normal_attr = generator.GenerateFuncAttribute(
attribute_name='age',
function=attrgenfunct.generate_normal_age,
parameters=[45, 30, 0, 130],
)
gender_city_comp_attr = generator.GenerateCateCateCompoundAttribute(
categorical1_attribute_name='gender',
categorical2_attribute_name='city',
lookup_file_name='gender-city-japanese.csv',
has_header_line=False,
unicode_encoding=unicode_encoding_used,
)
# -----------------------------------------------------------------------------
# Define how the generated records are to be corrupted (using methods from
# the corruptor.py module).
# For a value edit corruptor, the sum or the four probabilities given must
# be 1.0.
#
surname_misspell_corruptor = corruptor.CorruptCategoricalValue(
lookup_file_name='surname-misspell-japanese.csv',
has_header_line=False,
unicode_encoding=unicode_encoding_used,
)
edit_corruptor = corruptor.CorruptValueEdit(
position_function=corruptor.position_mod_normal,
char_set_funct=basefunctions.char_set_ascii,
insert_prob=0.0,
delete_prob=0.0,
substitute_prob=0.6,
transpose_prob=0.4,
)
missing_val_corruptor = corruptor.CorruptMissingValue()
# -----------------------------------------------------------------------------
# Define the attributes to be generated for this data set, and the data set
# itself.
#
attr_name_list = ['surname', 'age', 'gender', 'city', 'credit-card-number']
attr_data_list = [surname_attr, credit_card_attr, age_normal_attr, gender_city_comp_attr]
# Nothing to change here - set-up the data set generation object.
#
test_data_generator = generator.GenerateDataSet(
output_file_name=out_file_name,
write_header_line=True,
rec_id_attr_name=rec_id_attr_name,
number_of_records=num_org_rec,
attribute_name_list=attr_name_list,
attribute_data_list=attr_data_list,
unicode_encoding=unicode_encoding_used,
)
# Define the probability distribution of how likely an attribute will be
# selected for a modification.
# Each of the given probability values must be between 0 and 1, and the sum of
# them must be 1.0.
# If a probability is set to 0 for a certain attribute, then no modification
# will be applied on this attribute.
#
attr_mod_prob_dictionary = {
'surname': 0.5,
'age': 0.2,
'gender': 0.05,
'city': 0.05,
'credit-card-number': 0.2,
}
# Define the actual corruption (modification) methods that will be applied on
# the different attributes.
# For each attribute, the sum of probabilities given must sum to 1.0.
#
attr_mod_data_dictionary = {
'surname': [(0.9, surname_misspell_corruptor), (0.1, missing_val_corruptor)],
'age': [(0.1, missing_val_corruptor), (0.9, edit_corruptor)],
'gender': [(1.0, missing_val_corruptor)],
'city': [(1.0, missing_val_corruptor)],
'credit-card-number': [(0.1, missing_val_corruptor), (0.9, edit_corruptor)],
}
# Nothing to change here - set-up the data set corruption object
#
test_data_corruptor = corruptor.CorruptDataSet(
number_of_org_records=num_org_rec,
number_of_mod_records=num_dup_rec,
attribute_name_list=attr_name_list,
max_num_dup_per_rec=max_duplicate_per_record,
num_dup_dist=num_duplicates_distribution,
max_num_mod_per_attr=max_modification_per_attr,
num_mod_per_rec=num_modification_per_record,
attr_mod_prob_dict=attr_mod_prob_dictionary,
attr_mod_data_dict=attr_mod_data_dictionary,
)
# =============================================================================
# No need to change anything below here
# Start the generation process
#
rec_dict = test_data_generator.generate()
assert len(rec_dict) == num_org_rec # Check the number of generated records
# Corrupt (modify) the original records into duplicate records
#
rec_dict = test_data_corruptor.corrupt_records(rec_dict)
assert len(rec_dict) == num_org_rec + num_dup_rec # Check total number of records
# Write generate data into a file
#
test_data_generator.write()