def predictmatchflow(matchid,distancemetric = "dtw",direct=False,v=0,k=100,resultsfolder="results"):
folder = "direct" if direct else "indirect"
if not os.path.exists('../data/' + resultsfolder):
os.makedirs('../data/' + resultsfolder)
if not os.path.exists('../data/' + resultsfolder + "/" + folder):
os.makedirs('../data/' + resultsfolder +"/"+folder)
if not os.path.exists('../data/logs/' + resultsfolder):
os.makedirs('../data/logs/' + resultsfolder)
if not os.path.exists('../data/logs/' + resultsfolder +"/"+folder):
os.makedirs('../data/logs/' + resultsfolder+"/"+folder)
if distancemetric == "dtw":
dist = custom_dtw_distance
filename=str(matchid) + "_dtw"
else:
dist = naive_distance
filename= (str(matchid) + "_naive")
distancemetric = "naive"
print "Predicting flow of match %s with distancemetric %s and direct: %s and k: %s and resultsfolder: %s"%(matchid,distancemetric,direct,k,resultsfolder)
logdict = dict()
rows = c.execute("select id from match where not (id =?)",(matchid,)).fetchall()
trainset,windowstime = logger.executeandtime(lambda:getWindows([m for (m,) in rows]))
logdict["get windows"] = windowstime
if v > 0: print("windows retrieved")
knn,constructclassifiertime = logger.executeandtime(lambda:NearestNeighboursVP(
windows=trainset,
k=k,
weighted=True,
dist=dist))
logdict["construct classifier"] = constructclassifiertime
if v > 0: print("classifier constructed")
testset,testsettime = logger.executeandtime(lambda:getWindows([matchid]))
logdict["get testset"] = testsettime
if v > 0: print("testset retrieved")
def classify_match():
results = []
for window in testset:
result,classificationtime = logger.executeandtime(lambda:knn.predict_shotprobgoalprob(window,direct=direct))
(shotprobdom,shotprobother),(goalprobdom,goalprobother) = result
is_shot = 0
if window.is_shot():
is_shot = 1 if not window.is_defensive_error_shot() else -1
is_goal = 0
if window.is_goal():
is_goal = 1 if not window.is_defensive_error_goal() else -1
time = window.get_time()
half = window.matchhalf.halfid
dom_team = window.get_dominating_team()
fcb = 1 if dom_team == "32311" else 0 if dom_team == "None" else -1
results.append([half,time,fcb,shotprobdom,shotprobother,is_shot,
goalprobdom,goalprobother,is_goal,classificationtime])
np.savetxt('../data/' + resultsfolder + "/" + folder + "/" + filename,results)
tx = time //1000
if half ==2:
tx+= 45*60
if tx%(60*15) in range(0,5) and v > 0:
print "Time in-game:",logger.to_timestring(tx)
foo,classifytime = logger.executeandtime(classify_match)
logdict["match classification"] = classifytime
logfile = open('../data/logs/' + resultsfolder+ "/" + folder + "/" + filename,"w+")
json.dump(logdict,logfile)
print "Match", matchid, "classified and written to file", filename
new_model = True
resultsdir = "../../data/match_predictions/"
resultsfile = str(matchid) + "dtw10s_double_noweights.txt"
nnfun = lambda:NearestNeighboursVP(
windows=trainset,
k=100,
weighted=True,
dist=custom_dtw_distance)
if new_model and os.path.isfile(knnfile):
os.remove(knnfile)
if not os.path.isfile(knnfile):
rows = c.execute("select id from match where not (id =?)",(matchid,)).fetchall()
trainset = logger.execute(\
lambda:getWindows([m for (m,) in rows]),\
"Trainset built in")
knn = logger.execute(nnfun,"VP Tree built in")
pickle.dump(knn,open(knnfile,'wb'))
else:
#knn = pickle.load(open(knnfile,'rb'))
print("KNN succesfully loaded")
testset = logger.execute(lambda:getWindows([matchid]),\
"Testset built in")
def classify_windows():
results = []
t0 = pc()
for window in testset:
(shotprobdom,shotprobother),(goalprobdom,goalprobother) = \