SynthGen

Introduction

SynthGen is a string generator that generates name-like strings for synthesising name identifiers.
The main goal for developing SynthGen is to create strings that shares the characteristics of a known population.
This version of SynthGen generates low fidelity sex and ethnicity.

Users can tailor:

Total number of terms to generate: total_strings (int) (1,000)

characteristics related to gender/ethnicity

Proportion of names are unique (not correct yet): unique_percentage (int) (0.5)
Proportion of gender: gender_settings (dict) {'Male': 0.4, 'Female': 0.6}
Proportion of Strings overlapped by gender: gender_overlap_percentage (int) (0.15)
Proportion of ethnicity: ethnicity_settings (dict) {'White': 0.81, 'Asian': 0.096, 'Black': 0.042, 'Mixed': 0.03, 'Other': 0.022}
Proportion of Strings overlapped by ethnicity: ethnicity_overlap_percentages (dict) {('White', 'Black'): 0.1, ('White', 'Asian'): 0.05, ('White', 'Other'): 0.1}

String related characteristics:

string_settings = {
'lengths': [4, 5, 6, 7], # Specify lengths
'length_probabilities': [0.15, 0.3, 0.4, 0.15], # Specify length probabilities
'num_terms_range': [1, 2, 3], # Specify number of terms range
'num_terms_probabilities': [0.2, 0.6, 0.2] # Specify number of terms probabilities
}

Future direction - 1

Other string characteristics will be introduced to tailor, such as:

• apostrophes, hyphens, non-ISO characters
• vowel-consonant ratios

Uniqueness of terms will be a key goal to be retained, by gender and ethnicity.

Future direction - 2

To implement a insight tool to generate settings based on the loaded dataframe, and generate strings that match the characteristics of the loaded dataset.

Current work-around looks like:

uniqueness_requirements = { ('White', 'Male'): 0.85, ('White', 'Female'): 0.80, ('Black', 'Male'): 0.90, ('Black', 'Female'): 0.75, # Add more combinations as needed }

df = pd.read_csv('output_from_SynthGen.csv') grouped_df = df.groupby(['Ethnicity', 'Gender'])

for group, group_data in grouped_df: # Get the desired uniqueness for the current group from the dictionary desired_uniqueness = uniqueness_requirements.get(group, 0.8) # Default to 0.8 if not specified

# Calculate the number of unique names needed for the group based on desired uniqueness
total_names = len(group_data)
unique_names_needed = int(desired_uniqueness * total_names)

# Apply your sampling technique to ensure the desired level of uniqueness
# Replace this with your actual sampling logic
sampled_data = group_data.sample(unique_names_needed)

# Append the sampled data to the subsample DataFrame
subsample_df = subsample_df.append(sampled_data)

subsample_df.to_csv('subsampled_data.csv', index=False)