ensemble model and preprocessed files added

Codes/Models/ENSEMBLE.py

@@ -0,0 +1,146 @@
+# Ensemble for combining various models
+
+import cPickle
+import sys
+import operator
+import random
+import numpy as np
+from sklearn import linear_model
+from sklearn.metrics import mean_squared_error
+from sklearn.linear_model import Ridge
+
+class DocumentContainer(object):
+    def __init__(self, entity_pair, sentences, label,pos,l_dist,r_dist,entity_pos,ner,dep_d1,dep_d2,dep_path):
+        self.entity_pair = entity_pair
+        self.sentences = sentences 
+        self.label = label
+        self.pos_tags = pos
+        self.ner = ner
+        self.l_dist = l_dist
+        self.r_dist = r_dist
+        self.entity_pos = entity_pos
+        self.dep_dist1 = dep_d1
+        self.dep_dist2 =dep_d2
+        self.dep_path_bw_en = dep_path
+
+def read_data(file_name):
+	prob_data = cPickle.load(open(file_name))
+	# returning the probabilities for each class predicted, and the true label
+	return (prob_data[3],prob_data[2])
+
+class ensemble():
+
+	def train(self,classifier_probabilities,true_y):
+		# have a linear ensemble of classifiers which predicts the confirdence of each option separately and then choose the best option 
+		X=[]
+		y=[]
+		# construct X , by flatenning the probabilities. therefore a1,a2,a3,a4 and b1,b2,b3,b4 with y are flattened to a1,b1,I(y[0]) .. a2,b2,I(y[1]).. etc.
+		for i in range(len(classifier_probabilities[0])):
+			tempy=[0.0]*len(classifier_probabilities[0][0])
+			tempy[true_y[i][0]]=1.0
+			y.extend(tempy)
+			for j in range(len(classifier_probabilities[0][0])):
+				temp=[]
+				for k in range(len(classifier_probabilities)):
+					temp.append(classifier_probabilities[k][i][j])
+				X.append(temp)
+
+		print len(y),len(X),len(X[0]),'.... linear regression data details'
+		# linear model
+		regr = linear_model.LinearRegression()
+		regr.fit(np.array(X),np.array(y))
+		print 'Coefficients: \n', regr.coef_," ", type(regr.coef_)
+		return regr.coef_
+
+	def test(self,test_prob_classifier,weights,true_y,entity_pair):
+		for i in range(len(test_prob_classifier)):
+			test_prob_classifier[i]=weights[i]*test_prob_classifier[i]
+		y= (test_prob_classifier[0]+test_prob_classifier[1]+test_prob_classifier[2])/3
+		test_pr = self.pr(y,true_y,entity_pair)
+		return test_pr
+
+	def equal_test(self,classifier1,classifier2,classifier3,true_y,entity_pair):
+		y= (classifier1+classifier2+classifier3)/3
+		test_pr = self.pr(y,true_y,entity_pair)
+		return test_pr
+
+	def pr(self,predict_y,true_y,entity_pair):
+	    final_labels = []
+	    for label in true_y:
+	        if 0 in label and len(label) > 1:
+	            label = [x for x in label if x!=0]
+	        final_labels.append(label[:])
+
+	    total = 0
+	    for label in final_labels:
+	        if 0 in label:
+	            continue
+	        else:
+	            total += len(label)
+	    print "Total:",total
+	    results = []
+	    for i in range(predict_y.shape[0]):
+	        for j in range(1, predict_y.shape[1]):
+	            results.append([i,j,predict_y[i][j],entity_pair[i]])
+	    resultSorted = sorted(results, key=operator.itemgetter(2),reverse=True)
+	    p_p = 0.0
+	    p_n = 0.0
+	    n_p = 0.0
+	    pr = []
+	    prec = 0.0
+	    rec = 0.0
+	    p_p_final = 0.0
+	    p_n_final = 0.0
+	    n_p_final = 0.0
+	    prev = -1
+	    for g,item in enumerate(resultSorted):
+	        prev = item[2]
+	        if 0 in final_labels[item[0]]:
+	            if item[1] == 0:
+	                continue
+	            else:
+	                n_p += 1
+	        else:
+	            if item[1] in final_labels[item[0]]:
+	                p_p += 1
+	            else:
+	                p_n += 1
+
+	        pr.append([(p_p)/(p_p+n_p+p_n), (p_p)/total])
+	        if rec <= 0.3:
+	            prec = (p_p)/(p_p+n_p+p_n)
+	            rec = (p_p)/total
+	            p_p_final = p_p
+	            p_n_final = p_n
+	            n_p_final = n_p
+
+	    print "p_p:",p_p_final,"n_p:",n_p_final,"p_n:",p_n_final
+	    return [prec,rec,total,pr]
+
+if __name__=='__main__':
+
+	dataset_path = sys.argv[1]
+	# loading pre calculated probabilities of each class in the dev set
+	classifier_1_dev_prob = read_data(dataset_path+"/dev/prob_dev_WA.pkl")
+	classifier_2_dev_prob = read_data(dataset_path+"/dev/prob_dev_EA.pkl")
+	classifier_3_dev_prob = read_data(dataset_path+"/dev/prob_dev_PCNN.pkl")
+
+	# loading pre calculated probabilities of each class in the test set
+	classifier_1_test_prob = read_data(dataset_path+"/test/prob_WA.pkl")
+	classifier_2_test_prob = read_data(dataset_path+"/test/prob_EA.pkl")
+	classifier_3_test_prob = read_data(dataset_path+"/test/prob_PCNN.pkl")
+
+	# make entity_pairs
+	test_data_bags = cPickle.load(open(dataset_path+"/preprocessed_dataset/test.pkl"))
+	entity_pair=[data_bag.entity_pair for data_bag in test_data_bags]
+
+	ensemble_classifier = ensemble()
+	#test_pr = ensemble_classifier.equal_test(classifier_1_test_prob[0],classifier_2_test_prob[0],classifier_3_test_prob[0],classifier_1_test_prob[1],entity_pair)
+	weights = ensemble_classifier.train([classifier_1_dev_prob[0],classifier_2_dev_prob[0],classifier_3_dev_prob[0]],classifier_1_dev_prob[1])
+	test_pr = ensemble_classifier.test([classifier_1_test_prob[0],classifier_2_test_prob[0],classifier_3_test_prob[0]],weights,classifier_1_test_prob[1],entity_pair)
+
+	out = open(dataset_path+"/pr/pr.txt",'w')
+	for i in range(0,len(test_pr[3]),10):
+		res = test_pr[3][i]
+		out.write(str(res[0])+"\t"+str(res[1])+"\n")
+	out.close()