Predicting the Software Size and Effort from the Code Using Natural Language Processing
dataset = 'codesearchnet_auto.csv'
source = 'code'
target = 'cosmic_function_number'
num_labels = 4
batch_size = 64
epochs = 5
lr = 2e-5
dataset = 'green_balance_functions.csv'
source = 'Code'
target = 'Interaction' or 'Communication' or 'Process' or 'Effort'
num_labels = 1
batch_size = 16
epochs = 4
lr = 1e-4 or 2e-5
Metric | Performance |
---|---|
COSMIC size | ACC = 84.5% |
Interaction Event | NMAE = 1.10 |
Communication Event | NMAE = 1.80 |
Process Event | NMAE = 0.09 |
Total Event | NMAE = 0.13 |
Effort | NMAE = 0.18 |