def test_recommender(self):
# Run the recommender against the mock test data above and verify expected stats afterwards.
query = RecommenderQuery()
#query.queryItemIds = set();
#query.excludeItemIds = set();
#query.categoryIds = set();
#query.timeDeltaMax = None; # If set to one of the constants (DELTA_ZERO, DELTA_HOUR, etc.), will count item associations that occurred within that time delta as co-occurrent. If left blank, will just consider all items within a given patient as co-occurrent.
query.sortField = "tf"
query.limit = 16
# Go ahead and query for all since short list and can get expected calculation results for all
query.maxRecommendedId = 0
# Artificial constraint to focus only on test data
log.debug(
"Query with no item key input, just return ranks by general likelihood then."
)
headers = ["clinical_item_id", "score"]
expectedData = \
[ RowItemModel( [-2, 2.0/13], headers ),
RowItemModel( [-5, 2.0/13], headers ),
RowItemModel( [-6, 2.0/13], headers ),
RowItemModel( [-1, 1.0/13], headers ),
RowItemModel( [-3, 1.0/13], headers ),
RowItemModel( [-7, 1.0/13], headers ),
RowItemModel( [-8, 1.0/13], headers ),
RowItemModel( [-10,1.0/13], headers ),
RowItemModel( [-11,1.0/13], headers ),
RowItemModel( [-12,1.0/13], headers ),
RowItemModel( [-13,1.0/13], headers ),
RowItemModel( [-14,1.0/13], headers ),
RowItemModel( [-15,1.0/13], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug(
"Query with key item inputs for which no data exists. Effecitvely ignore it then, so just return ranks by general likelihood."
)
query.queryItemIds = set([-100])
expectedData = \
[ RowItemModel( [-2, 2.0/13], headers ),
RowItemModel( [-5, 2.0/13], headers ),
RowItemModel( [-6, 2.0/13], headers ),
RowItemModel( [-1, 1.0/13], headers ),
RowItemModel( [-3, 1.0/13], headers ),
RowItemModel( [-7, 1.0/13], headers ),
RowItemModel( [-8, 1.0/13], headers ),
RowItemModel( [-10,1.0/13], headers ),
RowItemModel( [-11,1.0/13], headers ),
RowItemModel( [-12,1.0/13], headers ),
RowItemModel( [-13,1.0/13], headers ),
RowItemModel( [-14,1.0/13], headers ),
RowItemModel( [-15,1.0/13], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug("Query with category filter on recommended results.")
query.queryItemIds = set([-100])
query.excludeCategoryIds = set([-1, -4, -5, -6])
expectedData = \
[ #RowItemModel( [-2, 2.0/13], headers ),
RowItemModel( [-5, 2.0/13], headers ),
RowItemModel( [-6, 2.0/13], headers ),
#RowItemModel( [-1, 1.0/13], headers ),
#RowItemModel( [-3, 1.0/13], headers ),
RowItemModel( [-7, 1.0/13], headers ),
RowItemModel( [-8, 1.0/13], headers ),
RowItemModel( [-10,1.0/13], headers ),
RowItemModel( [-11,1.0/13], headers ),
RowItemModel( [-12,1.0/13], headers ),
RowItemModel( [-13,1.0/13], headers ),
#RowItemModel( [-14,1.0/13], headers ),
#RowItemModel( [-15,1.0/13], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug(
"Query with category filter and specific exclusion filter on recommended results."
)
query.queryItemIds = set([-100])
query.excludeItemIds = set([-6, -10])
query.excludeCategoryIds = set([-1, -4, -5, -6])
expectedData = \
[ #RowItemModel( [-2, 2.0/13], headers ),
RowItemModel( [-5, 2.0/13], headers ),
#RowItemModel( [-6, 2.0/13], headers ),
#RowItemModel( [-1, 1.0/13], headers ),
#RowItemModel( [-3, 1.0/13], headers ),
RowItemModel( [-7, 1.0/13], headers ),
RowItemModel( [-8, 1.0/13], headers ),
#RowItemModel( [-10,1.0/13], headers ),
RowItemModel( [-11,1.0/13], headers ),
RowItemModel( [-12,1.0/13], headers ),
RowItemModel( [-13,1.0/13], headers ),
#RowItemModel( [-14,1.0/13], headers ),
#RowItemModel( [-15,1.0/13], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug(
"General query with a couple of input clinical items + one with no association data (should effectively be ignored)."
)
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set()
query.excludeCategoryIds = set()
expectedData = \
[ RowItemModel( [-6, (1.0/6)*(2.0/2)+(1.0/4)*(1.0/2)], headers ),
#RowItemModel( [-5, (1.0/6)*(2.0/2)+(1.0/4)*(1.0/2)], headers ),
#RowItemModel( [-2, (1.0/6)*(1.0/2)+(1.0/6)*(2.0/2)], headers ),
RowItemModel( [-3, (1.0/6)*(2.0/2)], headers ),
RowItemModel( [-7, (1.0/6)*(2.0/2)], headers ),
RowItemModel( [-8, (1.0/6)*(2.0/2)], headers ),
RowItemModel( [-14,(1.0/4)*(1.0/2)], headers ),
RowItemModel( [-15,(1.0/4)*(1.0/2)], headers ),
RowItemModel( [-1, (1.0/6)*(1.0/2)], headers ),
RowItemModel( [-10,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-11,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-12,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-13,(1.0/6)*(1.0/2)], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug("General query with category limit")
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set()
query.excludeCategoryIds = set([-2, -4, -5, -6])
expectedData = \
[ #RowItemModel( [-6, (1.0/6)*(2.0/2)+(1.0/4)*(1.0/2)], headers ),
#RowItemModel( [-5, (1.0/6)*(2.0/2)+(1.0/4)*(1.0/2)], headers ),
#RowItemModel( [-2, (1.0/6)*(1.0/2)+(1.0/6)*(2.0/2)], headers ),
RowItemModel( [-3, (1.0/6)*(2.0/2)], headers ),
#RowItemModel( [-7, (1.0/6)*(2.0/2)], headers ),
#RowItemModel( [-8, (1.0/6)*(2.0/2)], headers ),
#RowItemModel( [-14,(1.0/4)*(1.0/2)], headers ),
#RowItemModel( [-15,(1.0/4)*(1.0/2)], headers ),
RowItemModel( [-1, (1.0/6)*(1.0/2)], headers ),
RowItemModel( [-10,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-11,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-12,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-13,(1.0/6)*(1.0/2)], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug("General query with specific exclusion")
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set([-4, -3, -2])
query.excludeCategoryIds = set()
expectedData = \
[ RowItemModel( [-6, (1.0/6)*(2.0/2)+(1.0/4)*(1.0/2)], headers ),
#RowItemModel( [-5, (1.0/6)*(2.0/2)+(1.0/4)*(1.0/2)], headers ),
#RowItemModel( [-2, (1.0/6)*(1.0/2)+(1.0/6)*(2.0/2)], headers ),
#RowItemModel( [-3, (1.0/6)*(2.0/2)], headers ),
RowItemModel( [-7, (1.0/6)*(2.0/2)], headers ),
RowItemModel( [-8, (1.0/6)*(2.0/2)], headers ),
RowItemModel( [-14,(1.0/4)*(1.0/2)], headers ),
RowItemModel( [-15,(1.0/4)*(1.0/2)], headers ),
RowItemModel( [-1, (1.0/6)*(1.0/2)], headers ),
RowItemModel( [-10,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-11,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-12,(1.0/6)*(1.0/2)], headers ),
RowItemModel( [-13,(1.0/6)*(1.0/2)], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug("General query, sort by TF*IDF lift.")
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set()
query.excludeCategoryIds = set()
query.sortField = "lift"
expectedData = \
[ #RowItemModel( [-5, (13.0/2)*((1.0/6)*(2.0/2)+(1.0/4)*(1.0/2))], headers ),
#RowItemModel( [-2, (13.0/2)*((1.0/6)*(1.0/2)+(1.0/6)*(2.0/2))], headers ),
RowItemModel( [-3, (13.0/1)*((1.0/6)*(2.0/2))], headers ),
RowItemModel( [-7, (13.0/1)*((1.0/6)*(2.0/2))], headers ),
RowItemModel( [-8, (13.0/1)*((1.0/6)*(2.0/2))], headers ),
RowItemModel( [-6, (13.0/2)*((1.0/6)*(2.0/2)+(1.0/4)*(1.0/2))], headers ),
RowItemModel( [-14,(13.0/1)*((1.0/4)*(1.0/2))], headers ),
RowItemModel( [-15,(13.0/1)*((1.0/4)*(1.0/2))], headers ),
RowItemModel( [-1, (13.0/1)*((1.0/6)*(1.0/2))], headers ),
RowItemModel( [-10,(13.0/1)*((1.0/6)*(1.0/2))], headers ),
RowItemModel( [-11,(13.0/1)*((1.0/6)*(1.0/2))], headers ),
RowItemModel( [-12,(13.0/1)*((1.0/6)*(1.0/2))], headers ),
RowItemModel( [-13,(13.0/1)*((1.0/6)*(1.0/2))], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
def action_default(self):
"""Look for related orders by association / recommender methods"""
# If patient is specified then modify query and exclusion list based on items already ordered for patient
recentItemIds = set()
if self.requestData["sim_patient_id"]:
patientId = int(self.requestData["sim_patient_id"])
simTime = int(self.requestData["sim_time"])
# Track recent item IDs (orders, diagnoses, unlocked results, etc. that related order queries will be based off of)
manager = SimManager()
recentItemIds = manager.recentItemIds(patientId, simTime)
# Recommender Instance to test on
self.recommender = ItemAssociationRecommender()
self.recommender.dataManager.dataCache = webDataCache
# Allow caching of data for rapid successive queries
query = RecommenderQuery()
if self.requestData["sortField"] == "":
self.requestData["sortField"] = "P-YatesChi2-NegLog"
# P-Fisher-NegLog should yield better results, but beware, much longer to calculate
query.parseParams(self.requestData)
if len(query.excludeItemIds) == 0:
query.excludeItemIds = self.recommender.defaultExcludedClinicalItemIds(
)
if len(query.excludeCategoryIds) == 0:
query.excludeCategoryIds = self.recommender.defaultExcludedClinicalItemCategoryIds(
)
#query.fieldList.extend( ["prevalence","PPV","RR"] );
displayFields = list()
if self.requestData["displayFields"] != "":
displayFields = self.requestData["displayFields"].split(",")
# Exclude items already ordered for the patient from any recommended list
query.excludeItemIds.update(recentItemIds)
if not query.queryItemIds: # If no specific query items specified, then use the recent patient item IDs
query.queryItemIds.update(recentItemIds)
recommendedData = self.recommender(query)
if len(recommendedData) > 0:
# Denormalize results with links to clinical item descriptions
self.recommender.formatRecommenderResults(recommendedData)
# Display fields should append Format suffix to identify which version to display, but use original for header labels
(self.requestData["fieldHeaders"], displayFieldsFormatSuffixed
) = self.prepareDisplayHeaders(displayFields)
# Format for HTML and add a control field for interaction with the data
for dataModel in recommendedData:
self.prepareResultRow(dataModel, displayFields)
# Try organize by category
if self.requestData["groupByCategory"]:
recommendedData = self.recommender.organizeByCategory(
recommendedData)
colNames = ["controls"]
# "name" for code. ,"category_description"
colNames.extend(displayFieldsFormatSuffixed)
colNames.extend(["description"])
lastModel = None
htmlLines = list()
for dataModel in recommendedData:
newCategory = (lastModel is None
or lastModel["category_description"] !=
dataModel["category_description"])
showCategory = (self.requestData["groupByCategory"]
and newCategory)
# Limit category display if many repeats
if showCategory:
htmlLines.append(CATEGORY_HEADER_TEMPLATE % dataModel)
htmlLines.append(
self.formatRowHTML(dataModel, colNames, showCategory))
lastModel = dataModel
self.requestData["dataRows"] = str.join("\n", htmlLines)
def action_default(self):
"""Look for related orders by association / recommender methods"""
self.recommender = ItemAssociationRecommender()
# Instance to test on
self.recommender.dataManager.dataCache = webDataCache
query = RecommenderQuery()
if self.requestData["sortField"] == "":
self.requestData["sortField"] = "P-YatesChi2-NegLog"
# P-Fisher-NegLog should yield better results, but beware, much longer to calculate
query.parseParams(self.requestData)
if len(query.excludeItemIds) == 0:
query.excludeItemIds = self.recommender.defaultExcludedClinicalItemIds(
)
if len(query.excludeCategoryIds) == 0:
query.excludeCategoryIds = self.recommender.defaultExcludedClinicalItemCategoryIds(
)
#query.fieldList.extend( ["prevalence","PPV","RR"] );
displayFields = list()
if self.requestData["displayFields"] != "":
displayFields = self.requestData["displayFields"].split(",")
recommendedData = self.recommender(query)
if len(recommendedData) > 0:
# Denormalize results with links to clinical item descriptions
self.recommender.formatRecommenderResults(recommendedData)
# Display fields should append Format suffix to identify which version to display, but use original for header labels
(self.requestData["fieldHeaders"], displayFieldsFormatSuffixed
) = self.prepareDisplayHeaders(displayFields)
# Format for HTML and add a control field for interaction with the data
for dataModel in recommendedData:
self.prepareResultRow(dataModel, displayFields)
# Try organize by category
if self.requestData["groupByCategory"]:
recommendedData = self.recommender.organizeByCategory(
recommendedData)
colNames = ["controls"]
# "name" for code. ,"category_description"
colNames.extend(displayFieldsFormatSuffixed)
colNames.extend(["description"])
lastModel = None
htmlLines = list()
for dataModel in recommendedData:
newCategory = (lastModel is None
or lastModel["category_description"] !=
dataModel["category_description"])
showCategory = (self.requestData["groupByCategory"]
and newCategory)
# Limit category display if many repeats
if showCategory:
htmlLines.append(CATEGORY_HEADER_TEMPLATE % dataModel)
htmlLines.append(
self.formatRowHTML(dataModel, colNames, showCategory))
lastModel = dataModel
self.requestData["dataRows"] = str.join("\n", htmlLines)
def test_recommender(self):
# Run the recommender against the mock test data above and verify expected stats afterwards.
query = RecommenderQuery()
#query.queryItemIds = set();
#query.excludeItemIds = set();
#query.categoryIds = set();
#query.timeDeltaMax = None; # If set to one of the constants (DELTA_ZERO, DELTA_HOUR, etc.), will count item associations that occurred within that time delta as co-occurrent. If left blank, will just consider all items within a given patient as co-occurrent.
query.limit = 3
# Just get top 3 ranks for simplicity
query.maxRecommendedId = 0
# Artificial constraint to focus only on test data
log.debug(
"Query with no item key input, just return ranks by general likelihood then."
)
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-3], headers ),
RowItemModel( [-6], headers ),
RowItemModel( [-5], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug(
"Query with key item inputs for which no data exists. Effecitvely ignore it then, so just return ranks by general likelihood."
)
query.queryItemIds = set([-100])
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-3], headers ),
RowItemModel( [-6], headers ),
RowItemModel( [-5], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug("Query with category filter on recommended results.")
query.queryItemIds = set([-100])
query.excludeCategoryIds = set([-1, -4, -5, -6])
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-6], headers ),
RowItemModel( [-5], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug(
"Query with category filter and specific exclusion filter on recommended results."
)
query.queryItemIds = set([-100])
query.excludeItemIds = set([-6])
query.excludeCategoryIds = set([-1, -4, -5, -6])
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-5], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug(
"General query with a couple of input clinical items + one with no association data (should effectively be ignored)."
)
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set()
query.excludeCategoryIds = set()
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-4], headers ),
RowItemModel( [-6], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug(
"General query but set a limit on time delta worth counting item associations"
)
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set()
query.excludeCategoryIds = set()
query.timeDeltaMax = DELTA_HOUR
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-6], headers ),
RowItemModel( [-4], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug("General query with category limit")
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set()
query.excludeCategoryIds = set([-2, -4, -5, -6])
query.timeDeltaMax = DELTA_HOUR
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-4], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
log.debug("General query with specific exclusion")
query.queryItemIds = set([-2, -5, -100])
query.excludeItemIds = set([-4, -3, -2])
query.excludeCategoryIds = set()
query.timeDeltaMax = DELTA_HOUR
headers = ["clinical_item_id"]
expectedData = \
[ RowItemModel( [-6], headers ),
]
recommendedData = self.recommender(query)
self.assertEqualRecommendedData(expectedData, recommendedData, query)
# Keep track of each guideline name set
itemIdsByAdmitDxId = dict()
for admitDxId, sectionName, guidelineName, itemId, itemName, itemDescription, itemCount in resultsTable:
if admitDxId not in itemIdsByAdmitDxId:
itemIdsByAdmitDxId[admitDxId] = set()
itemIdsByAdmitDxId[admitDxId].add(itemId)
admitDxIdSectionGuidelineNameTuples.add(
(admitDxId, sectionName, guidelineName))
recommender = ItemAssociationRecommender()
for admitDxId, itemIds in itemIdsByAdmitDxId.iteritems():
print >> sys.stderr, admitDxId, len(itemIds)
recQuery = RecommenderQuery()
recQuery.excludeItemIds = recommender.defaultExcludedClinicalItemIds()
recQuery.excludeCategoryIds = recommender.defaultExcludedClinicalItemCategoryIds(
)
recQuery.queryItemIds = [admitDxId]
recQuery.timeDeltaMax = timedelta(1)
# Within one day
recQuery.countPrefix = "patient_"
recQuery.limit = TOP_ITEM_COUNT
# Top results by P-value
recQuery.sortField = "P-YatesChi2-NegLog"
results = recommender(recQuery)
#recommender.formatRecommenderResults(results);
for result in results:
itemIds.add(result["clinical_item_id"])
#print >> sys.stderr, result["description"];
print >> sys.stderr, admitDxId, len(itemIds)