def evaluation(u=None, d=None):
if u is None or d is None:
return render_template("evaluate_landing.html")
import numpy
department = app.db.query(Department).join(University) \
.filter(University.abbreviation==u) \
.filter(Department.abbreviation==d) \
.first()
if department is None:
abort(404) # department not found
# Retrieve the set of predicted and ground truth knowledge area labels
# for each course.
try:
knowledge_areas = {
'predicted': {
course.id:
predicted_knowledge_areas(course, result_set=g.result_set_raw)
for course in department.courses
},
'truth': {
course.id: ground_truth_knowledge_areas(course)
for course in department.courses
},
}
except RuntimeError:
# Return empty knowledge area lists if an error is encountered.
knowledge_areas = {
'predicted': {course.id: []
for course in department.courses},
'truth': {course.id: []
for course in department.courses},
}
# Calculate the jaccard coefficient and percentage correct of the
# prediction/truth sets, use these as 'correctness' metrics.
knowledge_areas['jaccard'] = {
course.id: float(
jaccard(knowledge_areas['predicted'][course.id],
knowledge_areas['truth'][course.id]))
for course in department.courses if knowledge_areas['truth'][course.id]
}
knowledge_areas['percent'] = {
course.id:
float(len(set(knowledge_areas['predicted'][course.id])\
.intersection(set(knowledge_areas['truth'][course.id])))\
/ len(knowledge_areas['truth'][course.id]))
for course in department.courses
if knowledge_areas['truth'][course.id]
}
return render_template(
"evaluate_department.html",
department=department,
knowledge_areas=knowledge_areas,
)
def evaluation(u=None, d=None):
if u is None or d is None:
return render_template("evaluate_landing.html")
import numpy
department = app.db.query(Department).join(University) \
.filter(University.abbreviation==u) \
.filter(Department.abbreviation==d) \
.first()
if department is None:
abort(404) # department not found
# Retrieve the set of predicted and ground truth knowledge area labels
# for each course.
try:
knowledge_areas = {
'predicted': {
course.id: predicted_knowledge_areas(
course,
result_set=g.result_set_raw)
for course in department.courses
},
'truth': {
course.id: ground_truth_knowledge_areas(course)
for course in department.courses
},
}
except RuntimeError:
# Return empty knowledge area lists if an error is encountered.
knowledge_areas = {
'predicted': {course.id: [] for course in department.courses},
'truth': {course.id: [] for course in department.courses},
}
# Calculate the jaccard coefficient and percentage correct of the
# prediction/truth sets, use these as 'correctness' metrics.
knowledge_areas['jaccard'] = {
course.id: float(jaccard(
knowledge_areas['predicted'][course.id],
knowledge_areas['truth'][course.id]
)) for course in department.courses
if knowledge_areas['truth'][course.id]
}
knowledge_areas['percent'] = {
course.id:
float(len(set(knowledge_areas['predicted'][course.id])\
.intersection(set(knowledge_areas['truth'][course.id])))\
/ len(knowledge_areas['truth'][course.id]))
for course in department.courses
if knowledge_areas['truth'][course.id]
}
return render_template("evaluate_department.html",
department=department,
knowledge_areas=knowledge_areas,)
def compare_departments(daid=None, dbid=None):
if None in [daid, dbid]:
departments = app.db.query(Department).all()
return render_template("compare_departments_landing.html",
departments=departments)
# Look up references to requested departments.
department_a = app.db.query(Department).get(daid)
department_b = app.db.query(Department).get(dbid)
# If either department isn't found, or if there is no result set
# (meaning no topics to infer) then simply 404.
if department_a is None or department_b is None or g.result_set_raw is None:
abort(404)
# Identify a set of topics for each department.
department_a_topics = set(topic_list(department_a, g.result_set_raw))
department_b_topics = set(topic_list(department_b, g.result_set_raw))
# Generate topic vectors for the two departments.
a_vector = topic_vector(department_a, g.result_set_raw)
b_vector = topic_vector(department_b, g.result_set_raw)
a_vector_string = a_vector.unpack(one=b'1', zero=b'0').decode('utf-8')
b_vector_string = b_vector.unpack(one=b'1', zero=b'0').decode('utf-8')
# Run similarity metrics.
similarity = dict()
similarity['jaccard'] = {
'name': 'Jaccard Index',
'range': '[0, 1]',
'description': 'Comparative set cardinality.',
'value': jaccard(department_a_topics, department_b_topics),
}
similarity['cosine'] = {
'name': 'Cosine Similarity',
'range': '[-1, 1]',
'description': 'Geometric cosine distance.',
'value': cosine_similarity(a_vector, b_vector),
}
similarity['euclidean'] = {
'name': 'Euclidean Distance',
'description': 'Geometric vector distance.',
'value': euclidean_distance(a_vector, b_vector),
}
# Remove common topics from the topic sets.
intersection = department_a_topics & department_b_topics
department_a_topics = department_a_topics - intersection
department_b_topics = department_b_topics - intersection
# Number of courses in each department.
num_courses_a = app.db.query(Course).join(Department) \
.filter(Department.id==daid).count()
num_courses_b = app.db.query(Course).join(Department) \
.filter(Department.id==dbid).count()
# Global list of departments for switching over.
departments = app.db.query(Department).all()
return render_template(
"compare_departments.html",
da=department_a,
db=department_b,
da_topics=department_a_topics,
db_topics=department_b_topics,
num_courses_a=num_courses_a,
num_courses_b=num_courses_b,
common_topics=intersection,
departments=departments,
similarity_metrics=similarity,
da_vector=a_vector_string,
db_vector=b_vector_string,
)
course.id: predicted_knowledge_areas(course, rs)
for course in gmu_cs.courses
},
'truth': {
course.id: ground_truth_knowledge_areas(course)
for course in gmu_cs.courses
},
} for rs in result_sets
]
print("Done.")
print("Calculate jaccard and percent metrics...")
for ka_dict in knowledge_areas:
ka_dict['jaccard'] = {
course.id: jaccard(
ka_dict['predicted'][course.id],
ka_dict['truth'][course.id]
) for course in gmu_cs.courses
if None not in [
ka_dict['predicted'][course.id],
ka_dict['truth'][course.id]
]
}
ka_dict['percent'] = {
course.id: len(set(ka_dict['predicted'][course.id])\
.intersection(set(ka_dict['truth'][course.id])))\
/ len(ka_dict['truth'][course.id])
for course in gmu_cs.courses
if ka_dict['truth'][course.id]
}
print("Done.")
def compare(depA, depB, rs):
dep_a_topics = topic_list(depA, rs)
dep_b_topics = topic_list(depB, rs)
return jaccard(dep_a_topics, dep_b_topics)
def compare_departments(daid=None, dbid=None):
if None in [daid, dbid]:
departments = app.db.query(Department).all()
return render_template("compare_departments_landing.html",
departments=departments)
# Look up references to requested departments.
department_a = app.db.query(Department).get(daid)
department_b = app.db.query(Department).get(dbid)
# If either department isn't found, or if there is no result set
# (meaning no topics to infer) then simply 404.
if department_a is None or department_b is None or g.result_set_raw is None:
abort(404)
# Identify a set of topics for each department.
department_a_topics = set(topic_list(department_a, g.result_set_raw))
department_b_topics = set(topic_list(department_b, g.result_set_raw))
# Generate topic vectors for the two departments.
a_vector = topic_vector(department_a, g.result_set_raw)
b_vector = topic_vector(department_b, g.result_set_raw)
a_vector_string = a_vector.unpack(one=b'1', zero=b'0').decode('utf-8')
b_vector_string = b_vector.unpack(one=b'1', zero=b'0').decode('utf-8')
# Run similarity metrics.
similarity = dict()
similarity['jaccard'] = {
'name': 'Jaccard Index',
'range': '[0, 1]',
'description': 'Comparative set cardinality.',
'value': jaccard(department_a_topics, department_b_topics),
}
similarity['cosine'] = {
'name': 'Cosine Similarity',
'range': '[-1, 1]',
'description': 'Geometric cosine distance.',
'value': cosine_similarity(a_vector, b_vector),
}
similarity['euclidean'] = {
'name': 'Euclidean Distance',
'description': 'Geometric vector distance.',
'value': euclidean_distance(a_vector, b_vector),
}
# Remove common topics from the topic sets.
intersection = department_a_topics & department_b_topics
department_a_topics = department_a_topics - intersection
department_b_topics = department_b_topics - intersection
# Number of courses in each department.
num_courses_a = app.db.query(Course).join(Department) \
.filter(Department.id==daid).count()
num_courses_b = app.db.query(Course).join(Department) \
.filter(Department.id==dbid).count()
# Global list of departments for switching over.
departments = app.db.query(Department).all()
return render_template("compare_departments.html",
da=department_a,
db=department_b,
da_topics=department_a_topics,
db_topics=department_b_topics,
num_courses_a=num_courses_a,
num_courses_b=num_courses_b,
common_topics=intersection,
departments=departments,
similarity_metrics=similarity,
da_vector=a_vector_string,
db_vector=b_vector_string,
)