def removeGerund(ehrChild):
separatedElementTag = Utilities.listFromCapitals(ehrChild);
if ( len(separatedElementTag) > 1 ):
taggedSeperatedElementTag = nltk.pos_tag(separatedElementTag);
separatedElementTag = [tag[0] for tag in taggedSeperatedElementTag if "VBG" not in tag[1]]
return "".join(separatedElementTag);
return ehrChild;
def compositeStringSimilarity(ehrClassField, fhirClassField, comparisonMethod, comparisonMethodArgs=[], highestResult=True, removeStopwords=True):
if ( comparisonMethod(ehrClassField, fhirClassField, *comparisonMethodArgs) == 1 ): return 1;
# If ehrClass string is composite, compare each word with the FHIR target using all of the metrics, and then use chosen combination method to produce a value, e.g. for each word, add these values, and then divide by number of words to get an average match across all words or return highest.
highestSimilarity = 0;
highestSimilarityWord = "";
totalSimilarity = 0;
ehrWords = Utilities.listFromCapitals(ehrClassField);
fhirWords = Utilities.listFromCapitals(fhirClassField);
if (removeStopwords): ehrWords = [word for word in ehrWords if word.lower() not in stopwords.words('english')];
for ehrWord in ehrWords:
highestSimilarityForEHRWord = 0;
for fhirWord in fhirWords:
similarity = comparisonMethod(ehrWord, fhirWord, *comparisonMethodArgs);
if ( similarity > highestSimilarity ):
highestSimilarity = similarity;
highestSimilarityWord = ehrWord;
if ( similarity > highestSimilarityForEHRWord ): highestSimilarityForEHRWord = similarity;
totalSimilarity += highestSimilarityForEHRWord;
if ( highestResult and len(highestSimilarityWord) > TranslationConstants.LENGTH_TO_IGNORE_IN_COMPOSITE_HIGHEST ):
return highestSimilarity;
else:
return old_div(totalSimilarity, max(float(len(ehrWords)), float(len(fhirWords))));
def getFHIRElements(root, classesToChildren, children=True, parents=True, recurse=True, selectiveRecurse=[], visited=[], addParentName=False, addStemmed=False, attributeTypeOverAttributeName=False, resolveFHIRReferences=False, otherFHIRClasses=None):
if len([excludedMatch for excludedMatch in TranslationConstants.EXCLUDED_FHIR_CLASS_TYPES if excludedMatch in root.__name__]): return [];
# Convert string to class, if not class.
if ( not inspect.isclass(root) ): root = eval(root);
# Ignore test classes.
if ( unittest.TestCase in inspect.getmro(root) or Exception in inspect.getmro(root) ): return [];
# Don't examine classes that don't use the 'elementsProperties' approach to list attributes.
if ( not callable(getattr(root, "elementProperties", None)) ): return [];
if ( root not in list(classesToChildren.keys()) ): classesToChildren[root] = set();
# Attributes of this class and parents.
attributes = root.elementProperties(root());
# List of parents (first element in tuple is this class).
parentList = inspect.getmro(root)[1:]
for parent in parentList:
if ( not callable(getattr(parent, "elementProperties", None)) ): continue;
# Don't reinclude attributes that are already included in a parent.
attributes = [item for item in attributes if item not in parent.elementProperties(parent())]
# If the type of an attribute is simply 'FHIRReference' we aim to resolve the scope of this reference by adding duplicate attributes for each potential reference type.
if resolveFHIRReferences:
newAttributes = [];
for attributeContainer in attributes:
if ( "FHIRReference" in attributeContainer[2].__name__ ):
sourceLines = inspect.getsource(root).split("\n");
for sourceLine in sourceLines:
if ( "self." + attributeContainer[0] in sourceLine ):
# If the list of possible references happens not to be two lines later, try three lines later.
if "FHIRReference" not in sourceLines[sourceLines.index(sourceLine) + 2]:
index = sourceLines.index(sourceLine) + 3;
else:
index = sourceLines.index(sourceLine) + 2;
for possibleFHIRReference in inspect.getsource(root).split("\n")[index].split("`")[3].split(","):
if possibleFHIRReference.strip() in [item.__name__ for item in otherFHIRClasses]:
attributeContainerAsList = list(attributeContainer);
attributeContainerAsList.insert(2, otherFHIRClasses[[item.__name__ for item in otherFHIRClasses].index(possibleFHIRReference.strip())]);
attributeContainer = tuple(attributeContainerAsList);
newAttributes.append(attributeContainer);
else:
newAttributes.append(attributeContainer);
attributes = newAttributes;
# For all attributes of this class (minus attributes of parent, which are typically generic).
for attributeContainer in attributes:
attribute = getattr(attributeContainer[2], "elementProperties", None)
attributeName = attributeContainer[0];
if len([excludedMatch for excludedMatch in TranslationConstants.EXCLUDED_FHIR_CLASS_TYPES if excludedMatch in attributeName]): continue
if children:
elementsOfChildren = TranslationUtilities.getFHIRElements(attributeContainer[2], {}, True, False, False);
# If a parent child (linked to another FHIR resource) has a suitable field that can hold data (e.g. value, text), then it doesn't matter if it's a parent, as it can effectively just act as a named container, so it might as well be a child (effective leaf node).
if not callable(attribute) or not set(TranslationConstants.FIELDS_THAT_INDICATE_RESOURCE_CAN_HOLD_ANY_DATA).isdisjoint(set(elementsOfChildren)):
TranslationUtilities.processAttribute(root, attributeTypeOverAttributeName, resolveFHIRReferences, classesToChildren, attributeContainer, attributeName);
# Add an additional pseudoelement with the parent name appended, which aims to represents the context given by the parent (e.g. child: given (not enough on its own) parent: HumanName full: givenHumanName (better representation of what is stored)).
if addParentName:
TranslationUtilities.processAttribute(root, attributeTypeOverAttributeName, resolveFHIRReferences, classesToChildren, attributeContainer, attributeName + str(root.__name__));
separatedParentName = Utilities.listFromCapitals(str(root.__name__));
# To reflect the fact that certain classes are better represented by removing the first word (e.g. HumanName -> Name)
if ( addStemmed and len(separatedParentName) > 1 ):
TranslationUtilities.processAttribute(root, attributeTypeOverAttributeName, resolveFHIRReferences, classesToChildren, attributeContainer, attributeName + separatedParentName[len(separatedParentName) - 1]);
if parents:
if callable(attribute):
TranslationUtilities.processAttribute(root, attributeTypeOverAttributeName, resolveFHIRReferences, classesToChildren, attributeContainer, attributeName);
# Don't expand from within FHIRReference, as it has a recursive reference to identifier (also doesn't appear to be captured correctly by the parser, e.g. organisation from Patient).
# Extensions classes appear in every class so don't show anything unique.
# Don't follow links to types that are of the root class itself.
# and attributeContainer[0] not in set([j for i in classesToChildren.values() for j in i])
if ( ( recurse and len(selectiveRecurse) == 0 ) or ( recurse and str(root.__name__) in selectiveRecurse ) or ( recurse and str(attributeContainer[2].__name__) in selectiveRecurse ) ) and callable(attribute) and "FHIRReference" not in str(root.__name__) and "Extension" not in str(attributeContainer[2]) and attributeContainer[2] != root and attributeContainer[0] not in visited:
visited.append(attributeContainer[0]);
TranslationUtilities.getFHIRElements(attributeContainer[2], classesToChildren, children, parents, recurse, selectiveRecurse, visited, addParentName, addStemmed, attributeTypeOverAttributeName, resolveFHIRReferences, otherFHIRClasses);
return classesToChildren;