kasrahabib / KM35NCDF

kasrahabib/KM35NCDF

This model is a fine-tuned version of sentence-transformers/all-MiniLM-L6-v2 on Software Requirements Dataset (SWARD) for classifying 19 Non-functional requirements. Note that based on literature, two out of 19 classes are Data and Behavior, belong to types of Functional software requirements. It achieves the following results on the evaluation set:

• Train Loss: 0.1691
• Validation Loss: 0.7548
• Epoch: 14
• Final Macro F1-score: 0.79

Labels : 0 or A -> Availability; 1 or AC -> Access Control; 2 or AU -> Audit; 3 or B -> Behaviour; 4 or D -> Data; 5 or FT -> Fault Tolerance; 6 or I -> Interface/Interoperability; 7 or LE -> Legal; 8 or LF -> Look and Feel; 9 or MN -> Maintainability; 10 or O -> Operational; 11 or PE -> Performance; 12 or PO -> Portability; 13 or RL -> Reliability; 14 or SA -> Safety; 15 or SC -> Scalability; 16 or SE -> Security; 17 or ST -> Stability; 18 or US -> Usability;

Usage Pipeline

from transformers import pipeline

frame_work = 'tf'
task = 'text-classification'
model_ckpt = 'kasrahabib/KM35NCDF '

software_requirment_cls = pipeline(task = task, model = model_ckpt, framework = frame_work)

example_1_US = 'Application needs to keep track of subtasks in a task.'
example_2_PE = 'The system shall allow users to enter time in several different formats.'
example_3_AC = 'The system shall allow users who hold any of the ORES/ORELSE/PROVIDER keys to be viewed as a clinical user and has full access privileges to all problem list options.'

software_requirment_cls([example_1_US, example_2_PE, example_3_AC])

[{'label': 'US', 'score': 0.9712953567504883},
 {'label': 'PE', 'score': 0.9457865953445435},
 {'label': 'AC', 'score': 0.9639136791229248}]

Model Inference:

import numpy as np
from transformers import AutoTokenizer, TFAutoModelForSequenceClassification

model_ckpt = 'kasrahabib/KM35NCDF '
tokenizer = AutoTokenizer.from_pretrained(model_ckpt)
model = TFAutoModelForSequenceClassification.from_pretrained(model_ckpt)

example_1_US = 'Application needs to keep track of subtasks in a task.'
example_2_PE = 'The system shall allow users to enter time in several different formats.'
example_3_AC = 'The system shall allow users who hold any of the ORES/ORELSE/PROVIDER keys to be viewed as a clinical user and has full access privileges to all problem list options.'
requirements = [example_1_US, example_2_PE, example_3_AC]

encoded_requirements = tokenizer(requirements, return_tensors = 'np', padding = 'longest')

y_pred = model(encoded_requirements).logits
classifications = np.argmax(y_pred, axis = 1)

classifications = [model.config.id2label[output] for output in classifications]
print(classifications)

['US', 'PE', 'AC']

Usage Locally Downloaded (e.g., GitHub):

1 - Clone the repository:

git lfs install
git clone url_of_repo

2 - Locate the path to the downloaded directory
3 - Write the link to the path in the model_ckpt variable

Then modify the code as below:

import numpy as np
from transformers import AutoTokenizer, TFAutoModelForSequenceClassification

model_ckpt =  'rest_of_the_path/KM35NCDF '
tokenizer = AutoTokenizer.from_pretrained(model_ckpt)
model = TFAutoModelForSequenceClassification.from_pretrained(model_ckpt)

example_1_US = 'Application needs to keep track of subtasks in a task.'
example_2_PE = 'The system shall allow users to enter time in several different formats.'
example_3_AC = 'The system shall allow users who hold any of the ORES/ORELSE/PROVIDER keys to be viewed as a clinical user and has full access privileges to all problem list options.'
requirements = [example_1_US, example_2_PE, example_3_AC]

encoded_requirements = tokenizer(requirements, return_tensors = 'np', padding = 'longest')

y_pred = model(encoded_requirements).logits
classifications = np.argmax(y_pred, axis = 1)

classifications = [model.config.id2label[output] for output in classifications]
print(classifications)

Training hyperparameters

The following hyperparameters were used during training:

• optimizer: {'name': 'Adam', 'weight_decay': None, 'clipnorm': None, 'global_clipnorm': None, 'clipvalue': None, 'use_ema': False, 'ema_momentum': 0.99, 'ema_overwrite_frequency': None, 'jit_compile': True, 'is_legacy_optimizer': False, 'learning_rate': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 2e-05, 'decay_steps': 6735, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}}, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False}
• training_precision: float32

Framework versions

• Transformers 4.26.1
• TensorFlow 2.11.0
• Datasets 2.10.0
• Tokenizers 0.13.2