save_location = '../../experiment_data/experiment_12'
k_list = [25, 50, 75, 100, 150, 200, 250, 300]
kernel_list = ["custom", "linear", "poly_2", "poly_3", "rbf"]
# results -> k -> kernal
results = []
limit_1 = 400 #training set
limit_2 = 400 #validation set
for k in k_list:
print "Running experiment for k=" + str(k) + "..."
ab = ABClassifier()
ab.download_cursors(limit_unlabeled=5000, limit_labeled=5000)
#ab.download_tweet_cursors(limit_unlabeled = 800, limit_labeled = 1000)
ab.run_lsa(k=k)
#pos_cursor_training = ab.labeled_collection.find({"bullying_label":"1"},timeout=False).limit(400)
#neg_cursor_training = ab.labeled_collection.find({"bullying_label":"0"},timeout=False).limit(400)
pos_cursor_training = ab.labeled_collection.find({
"bully": True
},
timeout=False).limit(400)
neg_cursor_training = ab.labeled_collection.find({
"bully": False
},
timeout=False).limit(400)
import sys
sys.path.append('..')
from ABClassifier.ABClassifier import ABClassifier
import numpy as np
ab = ABClassifier()
ab.download_cursors(limit_unlabeled = 1000, limit_labeled = 1000)
ab.run_lsa(k=100)
ab.compute_context_vectors()
pos_labeled_pws = ab.pairwise_similarity(ab.pos_labeled_cv_list)
neg_labeled_pws = ab.pairwise_similarity(ab.neg_labeled_cv_list)
unlabeled_pws = ab.pairwise_similarity(ab.unlabeled_cv_list)
print "done getting pws"
x = np.array(pos_labeled_pws.values())
a = np.asarray(x)
np.savetxt('pos_labeled.csv', a, delimiter=",")
y = np.array(neg_labeled_pws.values())
b = np.asarray(y)
np.savetxt('neg_labeled.csv', b, delimiter=",")
z = np.array(unlabeled_pws.values())
c = np.asarray(z)
np.savetxt('unlabeled.csv', c, delimiter=",")
save_location = '../../experiment_data/experiment_12'
k_list = [25, 50, 75, 100, 150, 200, 250, 300]
kernel_list = ["custom", "linear", "poly_2","poly_3", "rbf"]
# results -> k -> kernal
results = []
limit_1 = 400 #training set
limit_2 = 400 #validation set
for k in k_list:
print "Running experiment for k=" + str(k) + "..."
ab = ABClassifier()
ab.download_cursors(limit_unlabeled = 5000, limit_labeled = 5000)
#ab.download_tweet_cursors(limit_unlabeled = 800, limit_labeled = 1000)
ab.run_lsa(k=k)
#pos_cursor_training = ab.labeled_collection.find({"bullying_label":"1"},timeout=False).limit(400)
#neg_cursor_training = ab.labeled_collection.find({"bullying_label":"0"},timeout=False).limit(400)
pos_cursor_training = ab.labeled_collection.find({"bully":True},timeout=False).limit(400)
neg_cursor_training = ab.labeled_collection.find({"bully":False},timeout=False).limit(400)
training = []
tlabels = []
pos_validation = []
pos_vlabels = []
results_obj['neg_list_size'] = 0
results_obj['true_pos'] = 0
results_obj['true_neg'] = 0
results_obj['true_pos_rate'] = 0
results_obj['true_neg_rate'] = 0
results_obj['accuracy'] = 0
results_obj['num_iterations'] = 0
results[k] = results_obj
for i in xrange(0,20):
for k in k_list:
print "Running experiment for k = " + str(k)
ab = ABClassifier()
ab.download_cursors(limit_unlabeled = limit_1, limit_labeled = limit_1)
ab.run_lsa(k=k)
print "Starting classification..."
unlabeled_cursor = ab.db.tweets.find({"bullying_label" : {'$exists' :True}}, timeout = False)
pos_cursor = ab.labeled_collection.find({"bully":True},timeout=False).limit(unlabeled_cursor.count())
neg_cursor = ab.labeled_collection.find({"bully":False},timeout=False).limit(unlabeled_cursor.count())
unl = []
pos = []
neg = []
for u,p,n in zip(unlabeled_cursor, pos_cursor, neg_cursor):
results = {}
for t in thresholds:
results_obj = {}
results_obj['t'] = t
results_obj['pos_list_size'] = 0
results_obj['neg_list_size'] = 0
results_obj['true_pos'] = 0
results_obj['true_neg'] = 0
results_obj['true_pos_rate'] = 0
results_obj['true_neg_rate'] = 0
results_obj['accuracy'] = 0
results_obj['num_iterations'] = 0
results[int(t*100)] = results_obj
ab = ABClassifier()
ab.download_tweet_cursors(limit_unlabeled = 5000, limit_labeled = 5000)
ab.run_lsa(k=k)
context_tweets = [
"Literally go f**k yourself, because you're honestly pathetic.",
"f**k you f****n w***e go f**k yourself stupid bitch",
"but this bad I want to kick her ass cuz she thinks she's a hard chola like sit ur fat ass down lol",
"course he did he's a whipped bitch that will say anything to make u happy, unlike ur mum who called u fat",
"Fat pig. You're disgusting.",
"From some illiterate online keyboard warrior? Go back to sucking your butt buddy's fat junk.",
"God Says; Evil Don't Know The Way. You are gay with AIDS & your sin cost you your anointing! This is why you use DUST!"
"f**k you stupid f****t f*g"
]
tweet_cvs = []
results_obj['k'] = k
results_obj['pos_list_size'] = 0
results_obj['neg_list_size'] = 0
results_obj['true_pos'] = 0
results_obj['true_neg'] = 0
results_obj['true_pos_rate'] = 0
results_obj['true_neg_rate'] = 0
results_obj['accuracy'] = 0
results_obj['num_iterations'] = 0
results[k] = results_obj
for i in xrange(0, 20):
for k in k_list:
print "Running experiment for k = " + str(k)
ab = ABClassifier()
ab.download_cursors(limit_unlabeled=limit_1, limit_labeled=limit_1)
ab.run_lsa(k=k)
print "Starting classification..."
unlabeled_cursor = ab.db.tweets.find(
{"bullying_label": {
'$exists': True
}}, timeout=False)
pos_cursor = ab.labeled_collection.find({
"bully": True
}, timeout=False).limit(unlabeled_cursor.count())
neg_cursor = ab.labeled_collection.find({
"bully": False
},
thresholds = [0.5, 0.6, 0.7, 0.75, 0.8, 0.85]
results = {}
for t in thresholds:
results_obj = {}
results_obj['t'] = t
results_obj['pos_list_size'] = 0
results_obj['neg_list_size'] = 0
results_obj['true_pos'] = 0
results_obj['true_neg'] = 0
results_obj['true_pos_rate'] = 0
results_obj['true_neg_rate'] = 0
results_obj['accuracy'] = 0
results_obj['num_iterations'] = 0
results[int(t * 100)] = results_obj
ab = ABClassifier()
ab.download_cursors(limit_unlabeled=limit_1, limit_labeled=limit_1)
ab.run_lsa(k=k)
context_tweets = [
"Literally go f**k yourself, because you're honestly pathetic.",
"f**k you f****n w***e go f**k yourself stupid bitch",
"but this bad I want to kick her ass cuz she thinks she's a hard chola like sit ur fat ass down lol",
"course he did he's a whipped bitch that will say anything to make u happy, unlike ur mum who called u fat",
"Fat pig. You're disgusting.",
"From some illiterate online keyboard warrior? Go back to sucking your butt buddy's fat junk.",
"God Says; Evil Don't Know The Way. You are gay with AIDS & your sin cost you your anointing! This is why you use DUST!"
"f**k you stupid f****t f*g"
]
tweet_cvs = []
for c in context_tweets:
-Positive Examples, Negative Examples, Unlabeled
-Vary the number of input tweets to the Co-Occurrence Matrix
"""
import sys
sys.path.append('../..')
from ABClassifier.ABClassifier import ABClassifier
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
save_location = '../../experiment_data/experiment_2'
k_list = [5, 10, 25, 50, 100, 150, 250, 500]
for k in k_list:
ab = ABClassifier()
ab.download_cursors(limit_unlabeled=1000, limit_labeled=1000)
ab.run_lsa(k=k)
ab.compute_context_vectors(save_location=save_location)
print "Performing pairwise similarity measures..."
pos_labeled_pws = cosine_similarity(ab.pos_labeled_cv_list).flatten()
neg_labeled_pws = cosine_similarity(ab.neg_labeled_cv_list).flatten()
unlabeled_pws = cosine_similarity(ab.unlabeled_cv_list).flatten()
print "Done."
print "Saving..."
np.savetxt(save_location + '/pw_pos_' + str(k) + '.csv',
import sys
sys.path.append('..')
from ABClassifier.ABClassifier import ABClassifier
import numpy as np
ab = ABClassifier()
ab.download_cursors(limit_unlabeled=10000, limit_labeled=10000)
ab.run_lsa(k=100)
ab.compute_context_vectors()
pos_labeled_pws = ab.pairwise_similarity(ab.pos_labeled_cv_list)
neg_labeled_pws = ab.pairwise_similarity(ab.neg_labeled_cv_list)
unlabeled_pws = ab.pairwise_similarity(ab.unlabeled_cv_list)
print "done getting pws"
x = np.array(pos_labeled_pws.values())
a = np.asarray(x)
np.savetxt('pos_labeled.csv', a, delimiter=",")
y = np.array(neg_labeled_pws.values())
b = np.asarray(y)
np.savetxt('neg_labeled.csv', b, delimiter=",")
z = np.array(unlabeled_pws.values())
c = np.asarray(z)
np.savetxt('unlabeled.csv', c, delimiter=",")
-Positive Examples, Negative Examples, Unlabeled
-Vary the number of input tweets to the Co-Occurrence Matrix
-Uses ONLY twitter data for training/validation
"""
import sys
sys.path.append('../..')
from ABClassifier.ABClassifier import ABClassifier
import numpy as np
import os
save_location = '../../experiment_data/experiment_9'
ab = ABClassifier()
ab.download_tweet_cursors(limit_unlabeled=2500, limit_labeled=2500)
ab.run_lsa(k=150)
ab.compute_context_vectors(save_location=save_location)
print "Performing pairwise similarity measures..."
pos_labeled_pws = ab.pairwise_similarity(ab.pos_labeled_cv_list)
neg_labeled_pws = ab.pairwise_similarity(ab.neg_labeled_cv_list)
unlabeled_pws = ab.pairwise_similarity(ab.unlabeled_cv_list)
print "Done."
print "Saving..."
x = np.array(pos_labeled_pws.values())
"""
import sys
sys.path.append('../..')
from ABClassifier.ABClassifier import ABClassifier
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
save_location = '../../experiment_data/experiment_2'
k_list = [5, 10, 25, 50, 100, 150, 250, 500]
for k in k_list:
ab = ABClassifier()
ab.download_cursors(limit_unlabeled = 1000, limit_labeled = 1000)
ab.run_lsa(k=k)
ab.compute_context_vectors(save_location = save_location)
print "Performing pairwise similarity measures..."
pos_labeled_pws = cosine_similarity(ab.pos_labeled_cv_list).flatten()
neg_labeled_pws = cosine_similarity(ab.neg_labeled_cv_list).flatten()
unlabeled_pws = cosine_similarity(ab.unlabeled_cv_list).flatten()
print "Done."
print "Saving..."
np.savetxt(save_location + '/pw_pos_' + str(k) + '.csv', pos_labeled_pws, delimiter=",")
thresholds = [0.5, 0.6, 0.7, 0.75, 0.8, 0.85]
results = {}
for t in thresholds:
results_obj = {}
results_obj['t'] = t
results_obj['pos_list_size'] = 0
results_obj['neg_list_size'] = 0
results_obj['true_pos'] = 0
results_obj['true_neg'] = 0
results_obj['true_pos_rate'] = 0
results_obj['true_neg_rate'] = 0
results_obj['accuracy'] = 0
results_obj['num_iterations'] = 0
results[int(t*100)] = results_obj
ab = ABClassifier()
ab.download_tweet_cursors(limit_unlabeled = 2500, limit_labeled = 2500)
ab.run_lsa(k=k)
for i in xrange(0,1):
for t in thresholds:
print "Running experiment for t = " + str(t)
print "Starting classification..."
unlabeled_cursor = ab.db.tweets.find({"bullying_label" : {'$exists' :True}}, timeout = False).limit(limit_1)
pos_cursor = ab.labeled_collection.find({"bullying_label":"1"},timeout=False).skip(limit_1).limit(limit_2)
neg_cursor = ab.labeled_collection.find({"bullying_label":"0"},timeout=False).skip(limit_1).limit(limit_2)
thresholds = [0.5, 0.6, 0.7, 0.75, 0.8, 0.85]
results = {}
for t in thresholds:
results_obj = {}
results_obj['t'] = t
results_obj['pos_list_size'] = 0
results_obj['neg_list_size'] = 0
results_obj['true_pos'] = 0
results_obj['true_neg'] = 0
results_obj['true_pos_rate'] = 0
results_obj['true_neg_rate'] = 0
results_obj['accuracy'] = 0
results_obj['num_iterations'] = 0
results[int(t * 100)] = results_obj
ab = ABClassifier()
ab.download_tweet_cursors(limit_unlabeled=2500, limit_labeled=2500)
ab.run_lsa(k=k)
for i in xrange(0, 1):
for t in thresholds:
print "Running experiment for t = " + str(t)
print "Starting classification..."
unlabeled_cursor = ab.db.tweets.find(
{
"bullying_label": {
'$exists': True
}
}, timeout=False).limit(limit_1)
-Positive Examples, Negative Examples, Unlabeled
-Vary the number of input tweets to the Co-Occurrence Matrix
-Uses ONLY twitter data for training/validation
"""
import sys
sys.path.append('../..')
from ABClassifier.ABClassifier import ABClassifier
import numpy as np
import os
save_location = '../../experiment_data/experiment_9'
ab = ABClassifier()
ab.download_tweet_cursors(limit_unlabeled = 2500, limit_labeled = 2500)
ab.run_lsa(k=150)
ab.compute_context_vectors(save_location = save_location)
print "Performing pairwise similarity measures..."
pos_labeled_pws = ab.pairwise_similarity(ab.pos_labeled_cv_list)
neg_labeled_pws = ab.pairwise_similarity(ab.neg_labeled_cv_list)
unlabeled_pws = ab.pairwise_similarity(ab.unlabeled_cv_list)
print "Done."
print "Saving..."
