Combine repository specific labels with the universal model for issue…

… kind.

* To combine multiple models we define a base class IssueLabelModel
  that defines a common interface for all the models.

  * This way different models can just implement that interface and we
    can easily combine the results.

* UniveralKindLabelModel will be used for the generic model to predict
  label kind that is trained on all repositories.

  * The UniversalKindLabelModel class is based on the IssueLabeler code
    https://github.com/machine-learning-apps/Issue-Label-Bot/blob/536e8bf4928b03d522dd021c0464587747e90a87/flask_app/utils.py#L67

* kubeflow#70 Combine multiple models

py/label_microservice/models.py

@@ -0,0 +1,24 @@
+"""The models packages defines wrappers around different models."""
+
+class IssueLabelerModel:
+  """A base class for all Issue label models.
+
+  This class defines a common interface for all issue label models.
+  """
+
+  def predict_issue_labels(self, title:str , text:str ):
+    """Return a dictionary of label probabilities.
+
+    Args:
+      title: The title for the issue
+      text: The text for the issue
+
+    Return
+    ------
+    dict: Dictionary of label to probability of that label for the
+      the issue str -> float
+    """
+    raise NotImplementedError("predict_issue_probability should be overridden "
+                              "in a subclass.")
+
+

py/label_microservice/universal_kind_label_model.py

@@ -0,0 +1,71 @@
+
+from tensorflow.keras import models as keras_models
+from tensorflow.keras import utils as keras_utils
+
+import dill as dpickle
+
+from urllib.request import urlopen
+from label_microservice import models
+
+class UniversalKindLabelModel(models.IssueLabelerModel):
+  """UniversalKindLabelModel is a universal model that is trained across all repos.
+
+  The model predicts the kind for an issue.
+  """
+  def __init__(self,  class_names=['bug', 'feature', 'question']):
+    """Instantiate the model.
+
+    Args:
+      class_names: The specific label names to use for the three classes.
+    """
+    super(UniversalKindLabelModel, self).__init__()
+
+    # TODO(jlewi): We should probably parameterize the models rather than
+    # hardcoding it.
+    title_pp_url = "https://storage.googleapis.com/codenet/issue_labels/issue_label_model_files/title_pp.dpkl"
+    body_pp_url = 'https://storage.googleapis.com/codenet/issue_labels/issue_label_model_files/body_pp.dpkl'
+    model_url = 'https://storage.googleapis.com/codenet/issue_labels/issue_label_model_files/Issue_Label_v1_best_model.hdf5'
+    model_filename = 'downloaded_model.hdf5'
+
+    with urlopen(title_pp_url) as f:
+        self.title_pp = dpickle.load(f)
+
+    with urlopen(body_pp_url) as f:
+        self.body_pp = dpickle.load(f)
+
+    model_path = keras_utils.get_file(fname=model_filename, origin=model_url)
+    self.model = keras_models.load_model(model_path)
+
+    self.class_names = class_names
+
+  def predict_issue_labels(self, body:str, title:str):
+    """
+    Get probabilities for the each class.
+
+    Parameters
+    ----------
+    body: str
+       the issue body
+    title: str
+        the issue title
+
+    Returns
+    ------
+    Dict[str:float]
+
+    Example
+    -------
+    >>> issue_labeler = IssueLabeler(body_pp, title_pp, model)
+    >>> issue_labeler.get_probabilities('hello world', 'hello world')
+    {'bug': 0.08372017741203308,
+     'feature': 0.6401631832122803,
+     'question': 0.2761166989803314}
+    """
+    #transform raw text into array of ints
+    vec_body = self.body_pp.transform([body])
+    vec_title = self.title_pp.transform([title])
+
+    # get predictions
+    probs = self.model.predict(x=[vec_body, vec_title]).tolist()[0]
+
+    return {k:v for k,v in zip(self.class_names, probs)}

py/label_microservice/worker.py

@@ -223,6 +223,9 @@ def load_label_columns(self):
     def predict_issue_probability(self, repo_owner, repo_name, issue_num):
         """
         Predict probabilities of labels for an issue.
+
+        This function returns probabilities for all labels in the model.
+
         Args:
           repo_owner: repo owner
           repo_name: repo name
@@ -251,8 +254,11 @@ def predict_issue_probability(self, repo_owner, repo_name, issue_num):
         return label_probabilities, issue_embedding
 
     def predict_labels(self, repo_owner, repo_name, issue_num):
-        """
-        Predict labels for given issue.
+        """Predict labels for given issue.
+
+        This function only returns predictions for labels whose probability
+        exceeds the desired thresholds.
+
         Args:
           repo_owner: repo owner
           repo_name: repo name