Radon-defect-prediction-endpoints

This repository provised the APIs to expose Ansible and Tosca pre-trained defect prediction models.

Endpoints

• Get model: https://radon-defect-prediction.herokuapp.com/models
• Predict: https://radon-defect-prediction.herokuapp.com/predictions

Get model

Get a pre-trained model from the most similar project.

GET https://radon-defect-prediction.herokuapp.com/models?parm1=value1&...&paramN=valueN

Example: GET https://radon-defect-prediction.herokuapp.com/models?language=ansible&repository_size=560&comments_ratio=0.03&has_license=1

Parameters

Name	Value	Default
language	string	None. Choose between {ansible, tosca}
return_model	int [0: false, 1: true]	0
comments_ratio	numeric	0
commit_frequency	numeric	0
core_contributors	numeric	0
has_ci	[0: false, 1: true]	0
has_license	[0: false, 1: true]	0
iac_ratio	numeric	0
issue_frequency	numeric	0
repository_size	numeric	0

Return

A json containing the model id to use for calls to the predictions/ endpoint, a list of models, and a similarity score ([0-1]) between the client's and models' project.

{
    "model_id": 24242603,
    "similarity": 0.9999,
    "models": [
      {
        "type": "general"
        "rules": "|--- num_include <= 0.44\n|... "
      },
      {
        "type": "conditional"
        "rules": "... "
      },
    ]
}

If return_model=true, it returns the raw joblib model. Once saved somewhere, the model can be loaded and used in Python as follows:

import joblib

model = joblib.load('path/to/model.joblib', mmap_mode='r')
    
selected_features = model['selected_features']
normalizer = model['estimator'].named_steps['normalization']
tree_classifier = model['estimator'].named_steps['classification']

# selected_features can be used to reduce a new data frame to the same subset of model features
# normalizer (if not None) must be used to normalize the new data in the same fashion as model's training data
# tree_classifier can be used for predicting failure-prone script

Predict

Predict the failure-proneness of a file represented by a set of metrics.

GET https://radon-defect-prediction.herokuapp.com/predictions?parm1=value1&...&paramN=valueN

Example: GET https://radon-defect-prediction.herokuapp.com/predictions?language=ansible&model_id=24242603&num_names_with_vars=10&num_ignore_errors=3&num_conditions=1

Parameters

Name	Value	Default
language	string	None. Choose between {ansible, tosca}
model_id	int	None. It is mandatory. Use the model_id from the /models endpoint

The remaining parameters are the Ansible or Tosca metrics that can be extracted by the script. See how extract those metrics with AnsibleMetrics and ToscaMetrics.

Note: Metrics are passed instead of the raw script to avoid information disclosure.

Result A json file containing the prediction (i.e., failure_prone: true/false), and the decision that led to the prediction (absent if failure_prone: false).

{
  "failure_prone": true,
  "decision": [
    [
      "num_names_with_vars",
      ">",
      2.4
    ],
    [
      "num_ignore_errors",
      "<=",
      4.16
    ],
    [
      "num_conditions",
      "<=",
      17.12
    ],
    [
      "num_filters",
      "<=",
      0.0
    ]
  ]
}

In this example, a script has been predicted as failure-prone because:

num_names_with_vars > 2.4 AND num_ignore_errors <= 4.16 AND num_conditions <= 17.12 AND num_filters <= 0.