Generate random forest models with random parameters and cross-validate

rf_model.py

@@ -0,0 +1,62 @@
+from sklearn.ensemble import RandomForestClassifier
+import csv
+import random
+
+def clean_data(data):
+	# converts to floats and adds columns for differences
+	return [[float(v) for v in row] + [float(row[i])-float(row[i+10]) for i in range(0,10)] for row in data[1:]]
+
+data = []
+result = []
+with open('training_data.csv', 'rU') as csvfile:
+	rowreader = csv.reader(csvfile, delimiter=',')
+	for row in rowreader:
+		data.append(row[1:11]+row[12:-1])
+		result.append(row[-1])
+
+result=[0 if v=='FALSE' else 1 for v in result[1:]]		
+data=clean_data(data)
+
+# This is pretty silly, it creates random forests with 3% cross-validation and just looks for the one
+# that performs best on the cv data. This data seems to defy most methods.
+best_score=0.0
+best_model=None
+# This takes ages, but you can reduce the number of models it builds to make it faster
+for outer in range(500):
+	tr_data=[]
+	tr_result=[]
+	cv_data=[]
+	cv_result=[]
+	for i in range(0,len(data)):
+		if (random.random()<0.03):
+			cv_data.append(data[i])
+			cv_result.append(result[i])
+		else:
+			tr_data.append(data[i])
+			tr_result.append(result[i])
+
+	clf = RandomForestClassifier(n_estimators=random.randint(10,30), 
+								criterion="entropy" if random.random()<0.5 else "gini", 
+								max_features="auto" if random.random()<0.5 else None);
+	clf.fit(tr_data, tr_result)
+
+	match_count=0
+	for i in range(0, len(cv_data)):
+		if clf.predict(cv_data[i])[0]==cv_result[i]: match_count+=1
+
+	if (float(match_count)/len(cv_data))>best_score:
+		best_score=float(match_count)/len(cv_data)
+		best_model=clf
+
+test=[]
+with open('test_data.csv', 'rU') as csvfile:
+	rowreader = csv.reader(csvfile, delimiter=',')
+	for row in rowreader:
+		test.append(row)
+
+header=test[0]
+td=clean_data([row[1:11]+row[12:-1] for row in test])
+
+print ','.join(header)+',members_became_friends'
+for i in range(0,len(td)):
+	print ','.join(test[i+1])+('FALSE' if best_model.predict(td[i])[0]==0 else 'TRUE')