runner.py

import sys
import matplotlib.pyplot as plt
import project_functions as pf
import project_algorthims as pa
import monty
from cluster2d import cluster_svm as cluster
import numpy as np
import plottings

#test gating algorthim
if __name__ == "__main__":

	print "Reading Data..."
	sys_arguments = sys.argv
	users, batch_score, users_time, batch_time = pf.readData(sys_arguments[1])

	#run calculate_avg_score_per_batch on the global user scores
	global_batch = pf.calculate_avg_score_per_batch(batch_score)
	global_time = pf.calculate_avg_score_per_batch(batch_time)

	#Prep for creation of beta weights
	print "Working on prepping data for regression weights..."
	regressionData = pf.regressionDataPrep(global_time,global_batch, users, users_time)

	#this is the test parameter for a particular batch/project we are looking at so we exclude it from their aggregate score
	#It will default to the largest batch in the file, but can be set at the command line by using the word batch and then putting the batch number
	if len(sys_arguments) > 2:
		if sys_arguments[2] == "batch":
			mxBatch = int(sys_arguments[3])
			if mxBatch not in global_batch:
				print "there are no tasks in this batch, reverting to longest batch"
				mxBatch = pf.chooseBatch(batch_score)
		else:
			mxBatch = pf.chooseBatch(batch_score)
	else:
		mxBatch = pf.chooseBatch(batch_score)
	print '\033[1m' + "Current Batch: " +str(mxBatch) + '\033[0m'
	
	#Create a user dictionary based of of individual users and global scores
	scores, user_ten = pf.scoresLoop(users, users_time, mxBatch, global_batch, global_time)

	#Create a matrix from Scores Dictionary that has UserID, Past Performance, Time Performance, and Current Score for each user
	perfMat = pf.create_perf_arrays(scores)

	### Allow for parameter to loop through to calculate average profit function
	averagesum_2 = 0
	averagesum_3 = 0	
	profit_sum2 = 0
	profit_sum3 = 0
	if "run-loop" in sys_arguments:
		run = 30
		loop = True
	else:
		run = 1
		loop = False
	for i in xrange(run):
		#Calculate Centroids
		#initial_centroids = np.array(([-0.5,0.5],[0,0],[0.5,-0.5]))
		#initial_centroids2 = np.array(([-0.25,0.25],[0.25,-0.25]))
		centroids = cluster(perfMat[:,1],perfMat[:,2], 3, "Past Score (Normalized)", "Average Time (Normalized)", False)
		centroids2 = cluster(perfMat[:,1],perfMat[:,2], 2, "Past Score (Normalized)", "Average Time (Normalized)", False)
		print " "
		print '\033[1m' + "Centroid Cutoffs" '\033[0m'
		print "3 Cluster Centroids"
		print centroids
		print "2 Cluster Centroids"
		print centroids2

		#With Centroids as thresholds run each alogithm with the test parameters per users
		base = 200
		weights = {"Current":1}
		weights["Time"], weights["Past"] = pf.weightRegressions(regressionData)
		print " "
		print '\033[1m' + "Algorthim weights:" + '\033[0m'
		print "Time Weight:" + str(weights["Time"])
		print "Past Score Weight:" + str(weights["Past"])

		#With our weights and centroids compute every user's estimated number of questions before 
		#gold using each of our 6 algorthims
		questions = np.zeros((len(perfMat[:,1]),6))
		for u in range(len(perfMat[:,1])):
			alg_params = [perfMat[u,1], perfMat[u,3], perfMat[u,2], .98, centroids, weights, base]
			#Old Algorthims
			questions[u,0] = pa.StepWise(*alg_params)
			questions[u,1] = pa.StepWisePenalty(perfMat[u,1], perfMat[u,3], perfMat[u,2], .98, centroids, weights, 200, 0.5)
			questions[u,2] = pa.Attenuated(*alg_params)
			questions[u,3] = pa.AttenuatedContinous(*alg_params)

			#Attenuated Cluster Algorthims of Insterest
			questions[u,4] = pa.centroidThreshold(*alg_params)
			questions[u,5] = pa.centroidThreshold(perfMat[u,1], perfMat[u,3], perfMat[u,2], .98, centroids2, weights, base)
		
		#Now that we have questions before gold, this parameter calculates the profits from the first batch of work based on Samasource profit measurements. 
		questions_norm = [200]*len(questions[:,:])

		profit_3 = 0
		profit_2 = 0
		profit_norm = 0
		for i in range(len(questions[:,:])):
			profit_3 += (0.0348*questions[i,4])-((.0008*(questions[i,4])*22) + (.000225*questions[i,4])+(.0000066*questions[i,4]))
			profit_2 += (0.0348*questions[i,5])-((.0008*(questions[i,5])*22) + (.000225*questions[i,5])+(.0000066*questions[i,5]))
			profit_norm += (0.0348*questions_norm[i])-((.0008*(questions_norm[i])*22) + (.00045*questions_norm[i])+(.0000132*questions_norm[i]))
		
		profit_sum3 += profit_2
		profit_sum2 += profit_3
		
		print "Profit for 3 Clusters"
		print profit_3
		
		print "Profit for 2 Clusters"
		print profit_2
		
		print "Profit for Base Gating Algorithm"
		print profit_norm
		
		print " "
		print '\033[1m' + "How many questions have been added or subtracted"+ '\033[0m'
		print "StepWise StepWisePenalty Attenuated AttenuatedContinous 2Cluster 3Cluster"
		changeMatrix = questions - 200
		print changeMatrix

		average = np.mean(changeMatrix, axis=0)
		averagesum_2 += average[4]
		averagesum_3 += average[5]
		
		print " "
		print '\033[1m' + "Average Change in number of tasks before gold" + '\033[0m'
		print "StepWise StepWisePenalty Attenuated AttenuatedContinous 2Cluster 3Cluster"
		print average

	#Plot the results of the algorthim
	#plottings.score_centroid_distributions(perfMat[:,1], centroids, centroids2)
	#plottings.tenure_to_performance_plot(perfMat[:,1],user_ten, "Tenure", "Scores")
	plottings.scatterOfClusterResults(perfMat[:,1], perfMat[:,2], perfMat[:,3], questions, 'Scores', 'Times', 'Questions before Gold')

	## For a given project, once implemented the file would run the following function after the 
	#questionsCap = [50, 400]
	#secondRoundOutput = pf.secondRound(perfMat[:,1], newScores, questions[:,4:], centroids, centroids2, weights, questionsCap, perfMat)
	
	#Print Averages if run the full loop of average production
	if loop:
		print "Total Profit from 3 Cluster Algorithm: " + str(profit_sum3)
		print "Total Profit from 2 Cluster Algorithm: " + str(profit_sum2)
		print "Total Profit from Normal: " + str(profit_norm *30)
		
		print "Average Change before gold 2 centroids"
		print averagesum_2
		print "Average Change before gold 3 centroids"
		print averagesum_3
		
		print "Less Gold"
		lessGold = ((20/(averagesum_3+200))*300000)-((20/(200))*300000)
		print lessGold
		
		print "Gold Savings"
		goldSavings = lessGold*.03
		print goldSavings