def initialize_user(self, u, delete_user=True):
"""add user to the EventActionAggregate"""
self.users.append((u, delete_user))
for c in Category.objects.all():
try:
#See if event action aggregate exists.If it does, reset it to default.
eaa = EventActionAggregate.objects.get(user=u,category=c)
except:
#Else create one.
eaa = EventActionAggregate(user=u, category=c)
eaa.g, eaa.v, eaa.i, eaa.x = 0, 0, 0, 0
eaa.save()
def __init__(self, user=None, categories=None, num_categories=None):
"""
Initialize a new user and prepare to get recommendations from scratch.
If given categories, use those as preferred, otherwise if given number of
categories, create this many random ones
"""
if user:
self.user = user
self.delete_user = False
else:
# Create and assign a new user.
# if necessary later delete the user as well. But this might not be
# necessary since in Django we will be working with a fresh database
# that will in any case be destructed at the end of testing.
self.delete_user = True
success = False
count = 0
while not success:
count += 1
try:
self.user = User(username="******" + str(count),
password='******' + str(count))
self.user.save()
success = True
except:
success = False
for c in Category.objects.all():
try:
#See if event action aggregate exists.If it does, reset it to default.
eaa = EventActionAggregate.objects.get(user=self.user,
category=c)
except:
#Else create one.
eaa = EventActionAggregate(user=self.user, category=c)
eaa.g, eaa.v, eaa.i, eaa.x = 0, 0, 0, 0
eaa.save()
if categories:
categories = [self.get_category_id(c) for c in categories]
self.preferred_categories = set([c for c in categories])
else:
self.preferred_categories = set(
[c.id for c in self.get_random_categories(num_categories)])
def __init__(self, user=None, categories=None, num_categories=None):
"""
Initialize a new user and prepare to get recommendations from scratch.
If given categories, use those as preferred, otherwise if given number of
categories, create this many random ones
"""
if user:
self.user = user
self.delete_user = False
else:
# Create and assign a new user.
# if necessary later delete the user as well. But this might not be
# necessary since in Django we will be working with a fresh database
# that will in any case be destructed at the end of testing.
self.delete_user = True
success = False
count = 0
while not success:
count += 1
try:
self.user = User(username="******"+str(count),
password='******'+str(count))
self.user.save()
success = True
except:
success = False
for c in Category.objects.all():
try:
#See if event action aggregate exists.If it does, reset it to default.
eaa = EventActionAggregate.objects.get(user=self.user,category=c)
except:
#Else create one.
eaa = EventActionAggregate(user=self.user, category=c)
eaa.g, eaa.v, eaa.i, eaa.x = 0, 0, 0, 0
eaa.save()
if categories:
categories = [self.get_category_id(c) for c in categories]
self.preferred_categories = set([c for c in categories])
else:
self.preferred_categories = set([c.id for c in
self.get_random_categories(num_categories)])
def test_convergence(self):
categories = ml.recommend_categories(self.user)
#print "Categories: ", categories
picked_category = ml.sample_distribution(categories.items())[0]
#print "picked category: ", picked_category
picked_aggr = EventActionAggregate(user=self.user, category=picked_category)
lst = []
ctree = CachedCategoryTree()
parents = ctree.parents(picked_category)
count = 0
while count < 100:
count +=1
print "Round: %d\r"%count,
sys.stdout.flush()
# recommend a new set of categories
recommendation_scores = ml.recommend_categories(self.user)
cats = ml.sample_category_distribution(recommendation_scores.items(),
settings.N)
found_count = cats.count(picked_category)
#print "Categories: ",cats
#print "ID: ", picked_category.id
cats = set(cats)
cats.discard(picked_category.id)
# # G(oto) picked category
picked_aggr.g += found_count
picked_aggr.save()
# X all other categories
for c in cats:
if c in parents:
continue
try:
eaa = EventActionAggregate.objects.get(user=self.user, category=c)
except EventActionAggregate.DoesNotExist:
eaa = EventActionAggregate(user=self.user, category=c)
eaa.x += 1
eaa.save()
lst.append(found_count*100.0/settings.N)
plt.plot(lst,color="blue")
plt.title("Rate of learning one category")
plt.xlabel("Trials")
plt.ylabel("% of all Recommendations")
plt.savefig("learning/test_results/test.pdf")
plt.cla()
self.assertTrue(True)
def test_multi_category_recall(self,user=None):
if not user:
user = self.user
for c in Category.objects.all():
EventActionAggregate(user=user,category=c).save()
# Each color corresponds to the number of categories (position in list + 1) selected during the iteration.
colors = ["r","b","g","k","c","m","y"]
# k is the number of categories selected.
for k in range(1,8):
categories = set(ml.recommend_categories(user))
picked_categories = set(random.sample(categories,k))
print "User picked categories: ", picked_categories
# Generate mapping between categories and User's event action aggregate (GVIX store)
picked_cat_aggregates = dict([(c,EventActionAggregate.objects.get(user=user, category=c)) for c in picked_categories])
trials = count()
lst = []
while trials.next() < 50:
print "Loop: ", trials
mean_loop_count = count()
recall = 0
iterations = 10
G = {} #temporary dictionary that store G counts for selected categories during iterations.
while mean_loop_count.next() < iterations:
cats = set(ml.recommend_categories(user))
correct_recommendations = cats.intersection(picked_categories)
for c in correct_recommendations:
try:
G[c] += 1
except:
G[c] = 1
#picked_cat_aggregates[x].save()
cats.discard(picked_cat_aggregates[c])
#end of looping over correct_recommendations
recall += len(correct_recommendations)*100.0/len(picked_categories)
#end of looping over iterations to calculate means.
#####
#print "Recall: ", recall
#set and save the x values of discarded categories to the average number of times the category was G'd over iterations.
for key,value in G.iteritems():
picked_cat_aggregates[key].g += min(round(value/iterations),2)
picked_cat_aggregates[key].save()
#end of looping over temporary dictionary.
lst.append(recall*1.0/iterations)
#Variant 1: All items in the last iterations that were not in the picked category have been X'd once.
for c in cats:
try:
eaa = EventActionAggregate.objects.get(user=user, category=c)
except EventActionAggregate.DoesNotExist:
eaa = EventActionAggregate(user=user, category=c)
eaa.x += 1
eaa.save()
#end of adding X's
#end
print "Recall: ",lst
plt.plot(lst,color=colors[k-1],label=k)
for c in Category.objects.all():
#import pdb; pdb.set_trace()
try:
eaa = EventActionAggregate.objects.get(user=user, category=c)
eaa.g = 0
eaa.v = 0
eaa.i = 0
eaa.x = 0
eaa.save()
except:
pass
plt.title("Recall")
plt.xlabel("Trials")
plt.ylabel("% of User preferred categories")
#plt.legend()
plt.savefig("learning/test_results/recall.pdf")
plt.cla()
#import pdb; pdb.set_trace()
self.assertTrue(True)