def getEWMGamePerf(df, dates, i, window, com=0.3, metric='PTS_G'):
# get performance over window
datesWindow = getDatesWindow(dates, i, window)
windowVals = getWindowVals(df, datesWindow, metric)
# return ewm performance
return predictionMethods.applyEWMA(pd.Series(windowVals), param=com).values
def updateLabel(label,
dates,
i,
dfPlayer,
labelsDict,
metric='PTS_G',
train=True):
# unpack label info
labelType = list(labelsDict.keys())[0]
labelParams = list(labelsDict.values())[0]
# label = perf - mean(window)
if labelType == 'mean':
# get parameters
window = labelParams['window']
# get average performance over window
datesWindow = getDatesWindow(dates, i, window)
windowVals = getWindowVals(dfPlayer, datesWindow, metric)
labelShift = np.mean(windowVals)
# label = perf - ewm(window)[-1]
elif labelType == 'ewm':
# get parameters
window = labelParams['window']
com = labelParams['com']
# get most recent performance after applying ewm over window
datesWindow = getDatesWindow(dates, i, window)
windowVals = getWindowVals(dfPlayer, datesWindow, metric)
ewm_windowVals = predictionMethods.applyEWMA(pd.Series(windowVals),
param=com).values
labelShift = ewm_windowVals[-1]
# label = perf - prev_performance
elif labelType == 'raw':
# get previous performance
prevDate = dates[i - 1]
labelShift = getGamePerf(dfPlayer, prevDate, metric)
# label = perf
else:
labelShift = 0
return label - labelShift if train else label + labelShift
def topNTeammatesPerf(df, teammates, year, window, n=1, com=0.2, metric='PTS'):
perfList = np.array([])
for player in teammates:
# get teammates info
dfPlayer = df.loc[df.Player == player]
# get valid years
years = dfPlayer.loc[dfPlayer.Year < year, 'Year'].values
# get performance over window
yearsWindow = years[-window:]
windowVals = getWindowVals(dfPlayer, yearsWindow, metric)
# get ewm performance
ewmPerf = predictionMethods.applyEWMA(pd.Series(windowVals),
param=com).values
ewmPerf = ewmPerf[-1] if ewmPerf.size else 0
# append to performance list
perfList = np.append(perfList, ewmPerf)
return heapq.nlargest(n, perfList)
def updateLabel(label, years, i, df, labelsDict, metric='PTS', train=True):
# unpack label info
labelType = list(labelsDict.keys())[0]
labelParams = list(labelsDict.values())[0]
# label = perf - mean(window)
if labelType == 'mean':
# get parameters
window = labelParams['window']
# get performance over relevant years
yearsWindow = getYearsWindow(years, i, window)
windowVals = getWindowVals(df, yearsWindow, metric)
# compute mean
labelShift = np.mean(windowVals)
# label = perf - ewm(window)[-1]
elif labelType == 'ewm':
# unpack parameters
window = labelParams['window']
com = labelParams['com']
# get performance over relevant years
yearsWindow = getYearsWindow(years, i, window)
windowVals = getWindowVals(df, yearsWindow, metric)
# apply ewm and get most recent value
ewm_windowVals = predictionMethods.applyEWMA(pd.Series(windowVals),
param=com).values
labelShift = ewm_windowVals[-1]
# label = perf
else:
labelShift = 0
return label - labelShift if train else label + labelShift
def getFeature(dates,
i,
dfPlayer,
dfLeague,
teamsDict,
featuresDict,
metric='PTS_G'):
# initialize
featureVec = np.array([])
num_hsvt_features = 0
for feature, featureParams in featuresDict.items():
# compute standard deviation (std) of window values
if feature == 'std':
# get parameters
window = featureParams['window']
# compute standard deviation of performance over relevant dates
datesWindow = dates[i - window:i]
windowVals = getWindowVals(dfPlayer, datesWindow, metric)
featureVal = np.std(windowVals)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# compute mean of window values
if feature == 'mean':
# get parameters
window = featureParams['window']
# compute average performance over relevant dates
datesWindow = dates[i - window:i]
windowVals = getWindowVals(dfPlayer, datesWindow, metric)
featureVal = np.mean(windowVals)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get performance of most recent game after applying ewm on window values
if feature == 'ewm':
# get parameters
window = featureParams['window']
com = featureParams['com']
# compute performance of relevant dates & apply ewm
datesWindow = dates[i - window:i]
windowVals = getWindowVals(dfPlayer, datesWindow, metric)
ewm_windowVals = predictionMethods.applyEWMA(pd.Series(windowVals),
param=com).values
# get most recent game performance
featureVal = ewm_windowVals[-1]
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get opponent information
if feature == 'oppt':
# get parameters
window = featureParams['window']
# get opponent on game date
currDate = dates[i]
oppTeam = dfPlayer.loc[dfPlayer.gmDate == currDate,
'opptAbbr'].values[0]
# get average performance of opposing teams allowed by oppTeam & league
oppTeamOppPerf = getTeamOppPerf(teamsDict, oppTeam, currDate)
leagueOppPerf = getLeagueOppPerf(teamsDict, currDate)
featureVal = oppTeamOppPerf / leagueOppPerf
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get team location on game date
if feature == 'teamLoc':
# get team location on game date
currDate = dates[i]
featureVal = getTeamLoc(dfPlayer, currDate)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
# get outcome of most recent game after applying ewm on window values
if feature == 'gmOutcome':
# get parameters
window = featureParams['window']
com = featureParams['com']
# compute outcomes of games during dates window & apply ewma
datesWindow = dates[i - window:i]
gmOutcomes = getGameOutcomes(dfPlayer, datesWindow)
gmOutcomes = predictionMethods.applyEWMA(pd.Series(gmOutcomes),
param=com).values
# get most recent game outcome
featureVal = gmOutcomes[-1]
# append to feature vector
featureVec = np.append(featureVec, featureVal)
# get teammate information
#if feature == 'teammates':
return featureVec, num_hsvt_features
def stats(player, dfLeague, teamsDict, window, oppWindow, com):
dfPlayer = dfLeague[dfLeague.Player == player]
dates = dfPlayer.gmDate.values
currPerfs = np.array([])
deltaPerfs = np.array([])
prevPerfs = np.array([])
prev_mean = np.array([])
prev_prev = np.array([])
prevGmResults = np.array([])
stds = np.array([])
teamLocs = np.array([])
y_list = np.array([])
y_mean_list = np.array([])
mean_list = np.array([])
opps = np.array([])
x_means = np.array([])
for i in range(2, len(dates)):
currDate = dates[i]
prevDate = dates[i - 1]
prevprevDate = dates[i - 2]
datesWindow = getDatesWindow(dates, i, window)
# get perfomance values
currPerf = getGamePerf(dfPlayer, currDate)
prevPerf = getGamePerf(dfPlayer, prevDate)
deltaPerf = currPerf - prevPerf
#windowVals = getWindowVals(dfPlayer, datesWindow)
#deltaPerf = currPerf - predictionMethods.applyEWMA(pd.Series(windowVals), param=com).values[-1]
meanPerf = np.mean(getSeasonPerf(dfPlayer, prevprevDate))
prevprevPerf = getGamePerf(dfPlayer, prevprevDate)
# game outcome
gmResults = getGameOutcomes(dfPlayer, datesWindow)
gmResults = predictionMethods.applyEWMA(pd.Series(gmResults),
param=com).values
prevGmResult = gmResults[-1]
# std
windowVals = getWindowVals(dfPlayer, datesWindow)
std = np.std(windowVals)
# mean
x_mean = np.mean(windowVals) if windowVals.size else 0
# team location
teamLoc = getTeamLoc(dfPlayer, currDate)
# teammate effect
team = getTeam(dfPlayer, currDate)
teammates = getTeammates(dfLeague, player, team, currDate)
y, y_mean, mean = teammateEffect(dfLeague, teammates, currDate, window,
1, com)
# opponent effect (position invariant)
"""position = dfPlayer.loc[dfPlayer.gmDate == currDate, 'playPos'].values[0]
oppTeam = getOppTeam(dfPlayer, currDate)
oppTeamOppPerf = getTeamOppPosPerf(teamsDict, oppTeam, currDate, position)
leagueOppPerf = getLeagueOppPosPerf(teamsDict, currDate, position, oppWindow)
opp = oppTeamOppPerf / leagueOppPerf if leagueOppPerf else 1"""
# append
x_means = np.append(x_means, x_mean)
currPerfs = np.append(currPerfs, currPerf)
prevPerfs = np.append(prevPerfs, prevPerf)
deltaPerfs = np.append(deltaPerfs, deltaPerf)
prev_mean = np.append(prev_mean, prevPerf - meanPerf)
prev_prev = np.append(prev_prev, prevPerf - prevprevPerf)
prevGmResults = np.append(prevGmResults, prevGmResult)
stds = np.append(stds, std)
teamLocs = np.append(teamLocs, teamLoc)
y_list = np.append(y_list, y)
y_mean_list = np.append(y_mean_list, y_mean)
mean_list = np.append(mean_list, mean)
df = pd.DataFrame({
'gmDate': dates[2:],
'PTS': currPerfs,
'Delta': deltaPerfs,
'mean(X(t-1))': x_means,
'Prev': prevPerfs,
'std': stds,
'Prev-Mean': prev_mean,
'Prev-Prev': prev_prev,
'PrevGmRslt': prevGmResults,
'teamLoc': teamLocs,
'Y(t-1)': y_list,
'Y(t-1)-mean(Y(t-2))': y_mean_list,
'mean(Y(t-1))': mean_list
})
return df
def getFeature(years, i, dfPlayer, dfLeague, featuresDict, metric='PTS'):
featureVec = np.array([])
for feature, featureParams in featuresDict.items():
if feature == 'std':
# unpack parameters
window = featureParams['window']
# get performance over relevant years
yearsWindow = getYearsWindow(years, i, window)
windowVals = getWindowVals(dfPlayer, yearsWindow, metric)
# compute standard deviation
featureVal = np.std(windowVals)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
if feature == 'mean':
# unpack parameters
window = featureParams['window']
# get performance over relevant years
yearsWindow = getYearsWindow(years, i, window)
windowVals = getWindowVals(dfPlayer, yearsWindow, metric)
# compute mean
featureVal = np.mean(windowVals)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
if feature == 'ewm':
# unpack parameters
window = featureParams['window']
com = featureParams['com']
# get performance over relevant years
yearsWindow = getYearsWindow(years, i, window)
windowVals = getWindowVals(dfPlayer, yearsWindow, metric)
# apply ewm and get most recent value
ewm_windowVals = predictionMethods.applyEWMA(pd.Series(windowVals),
param=com).values
featureVal = ewm_windowVals[-1]
# append to feature vector
featureVec = np.append(featureVec, featureVal)
if feature == 'teammates':
# unpack parameters
window = featureParams['window']
com = featureParams['com']
n = featureParams['n']
# get current year and player
currYear = years[i]
player = dfPlayer.Player.values[0]
# get current team and teammates
team = getTeam(dfPlayer, currYear)
teammates = getTeammates(dfLeague, player, team, currYear)
featureVal = topNTeammatesPerf(dfLeague, teammates, currYear,
window, n, com, metric)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
return featureVec
def getFeature(dates, i, dfPlayer, dfLeague, teamsDict, featuresDict, metric='PTS_G'):
# initialize
featureVec = np.array([])
num_hsvt_features = 0
for feature, featureParams in featuresDict.items():
# compute standard deviation (std) of window values
if feature == 'std':
# get parameters
window = featureParams['window']
# compute standard deviation of performance over relevant dates
datesWindow = getDatesWindow(dates, i, window)
windowVals = getWindowVals(dfPlayer, datesWindow, metric)
featureVal = np.std(windowVals) if windowVals.size else 0
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# compute mean of window values
if feature == 'mean':
# get parameters
window = featureParams['window']
# compute average performance over relevant dates
datesWindow = getDatesWindow(dates, i, window)
windowVals = getWindowVals(dfPlayer, datesWindow, metric)
featureVal = np.mean(windowVals) if windowVals.size else 0
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get performance of most recent game after applying ewm on window values
if feature == 'ewm':
# get parameters
window = featureParams['window']
com = featureParams['com']
# compute performance of relevant dates & apply ewm
ewm_windowVals = getEWMGamePerf(dfPlayer, dates, i, window, com=com, metric='PTS_G')
# get most recent game performance
featureVal = ewm_windowVals[-1]
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get opponent information
if feature == 'oppt':
# get parameters
window = featureParams['window']
# get opponent on game date
currDate = dates[i]
oppTeam = getOppTeam(dfPlayer, currDate)
# get average performance of opposing teams allowed by oppTeam & league
oppTeamOppPerf = getTeamOppPerf(teamsDict, oppTeam, currDate, window)
leagueOppPerf = getLeagueOppPerf(teamsDict, currDate, window)
# if leagueOppPerf = 0 (no games played yet) then set featureVal = 1
featureVal = oppTeamOppPerf / leagueOppPerf if leagueOppPerf else 1
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get outcome of most recent game after applying ewm on window values
if feature == 'gmOutcome':
# get parameters
window = featureParams['window']
com = featureParams['com']
# compute outcomes of games during dates window & apply ewma
datesWindow = getDatesWindow(dates, i, window)
gmOutcomes = getGameOutcomes(dfPlayer, datesWindow)
gmOutcomes = predictionMethods.applyEWMA(pd.Series(gmOutcomes), param=com).values
# get most recent game outcome
featureVal = gmOutcomes[-1]
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get teammate information
if feature == 'teammates':
# get parameters
window = featureParams['window']
com = featureParams['com']
n = featureParams['n']
# get current date and player
currDate = dates[i]
player = dfPlayer.Player.values[0]
# get current team and teammates
team = getTeam(dfPlayer, currDate)
teammates = getTeammates(dfLeague, player, team, currDate)
featureVal = topNTeammatesPerf(dfLeague, teammates, currDate, window, n, com, metric)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += n
# get delta performance between previous game and prior performances
if feature == 'delta':
# get parameters
window = featureParams['window']
com = featureParams['com']
# get previous performance
prevDate = dates[i-1]
prevGmPerf = getGamePerf(dfPlayer, prevDate)
# get performance over window of games prior to previous date
#datesWindow = getDatesWindow(dates, i-1, window)
#windowVals = getWindowVals(dfPlayer, datesWindow, metric)
#ewm_windowVals = predictionMethods.applyEWMA(pd.Series(windowVals), param=com).values
#featureVal = prevGmPerf - ewm_windowVals[-1]
# get difference in performances
prev_prevDate = dates[i-2]
prev_prevGmPerf = getGamePerf(dfPlayer, prev_prevDate)
featureVal = prevGmPerf - prev_prevGmPerf
# append to feature vector
featureVec = np.append(featureVec, featureVal)
num_hsvt_features += 1
# get team location on game date
if feature == 'teamLoc':
# get team location on game date
currDate = dates[i]
featureVal = getTeamLoc(dfPlayer, currDate)
# append to feature vector
featureVec = np.append(featureVec, featureVal)
return featureVec, num_hsvt_features
