def calculateProbsOfEachGameInASeason(csv_file_name, predict):
data, indexToTeam, teamToIndex, indexToGamesPlayed = epl.getData(
csv_file_name)
game_data = []
for i in range(len(data["Home Team"])):
homeTeam = data["Home Team"][i]
awayTeam = data["Away Team"][i]
homeIndex = teamToIndex[homeTeam]
awayIndex = teamToIndex[awayTeam]
winnerProp, tieProb, loserProb = predict(homeIndex, awayIndex, i)
total = winnerProp + tieProb + loserProb
winnerProp = winnerProp / total
tieProb = tieProb / total
loserProb = loserProb / total
game_data.append({
'Home Team': homeTeam,
'Away Team': awayTeam,
'Round Number': int(data['Round Number'][i]),
'Date': data['Date'][i],
'Location': data['Location'][i],
'home_win': float(winnerProp),
'tie': float(tieProb),
'away_win': float(loserProb),
})
return game_data
def createPredictGameFunction(filename):
inputs, labels = cdata.getAllMyData([filename], True, 0)
data, indexToTeam, teamToIndex, indexToGamesPlayed = epl.getData(filename)
model = makeModel(inputs, inputs.shape[1], labels.shape[1])
outputs = model.predict(inputs)
def predict(homeIndex, awayIndex, index):
output = np.array(outputs[index])
p_output = output / output.sum()
return (p_output[0], p_output[2], p_output[1])
return predict
def main(csv_file_name='epl-2021.csv',
output_file_name='epl-predictions-stats.json'):
model = epl.getData(csv_file_name)
data, indexToTeam, teamToIndex, indexToGamesPlayed = model
predictor = predictions_tensorflow.createPredictGameFunction(csv_file_name)
indexToPlaceFinishedToTimesFinished, indexToMeanPoints, indexToTeam = probs.calculateProbs(
10**5, csv_file_name, predictor)
week = 40
A = np.array(epl.getMatrixForSeason(week, model, None))
R = pagerank.rank(A)
rankings = [(indexToMeanPoints[i], i)
for i in range(len(indexToMeanPoints))]
rankings.sort(reverse=True)
jsonData = []
for value in rankings:
probabilties = indexToPlaceFinishedToTimesFinished[value[1]]
jsonData.append({
"name":
indexToTeam[value[1]],
"probability":
list(probabilties),
"expected":
value[0],
"championslegue":
probabilties[0] + probabilties[1] + probabilties[2] +
probabilties[3],
"relegated":
probabilties[-1] + probabilties[-2] + probabilties[-3],
"pagerank":
R[value[1]][0],
})
games_data = probs.calculateProbsOfEachGameInASeason(
csv_file_name, predictor)
with open(output_file_name, 'w') as out:
out.write(json.dumps({'teams': list(jsonData), 'games': games_data}))
def getAllMyData(
filenames=[
#"old_data/american-football.csv/nfl-2020.csv",
#"old_data/american-football.csv/nfl-2019.csv",
#"old_data/american-football.csv/nfl-2018.csv",
#"old_data/american-football.csv/nfl-2017.csv",
#"old_data/american-football.csv/nfl-2016.csv",
"old_data/bundesliga-2018.csv",
"old_data/bundesliga-2019.csv",
"old_data/bundesliga-2020.csv",
"old_data/efl-championship-2019.csv",
"old_data/efl-championship-2020.csv",
"old_data/la-liga-2018.csv",
"old_data/la-liga-2019.csv",
#"old_data/la-liga-2020.csv",
"old_data/ligue-1-2018.csv",
"old_data/ligue-1-2019.csv",
"old_data/ligue-1-2020.csv",
"old_data/serie-a-2017.csv",
"old_data/serie-a-2018.csv",
"old_data/serie-a-2019.csv",
"old_data/serie-a-2020.csv",
"old_data/turkey-super-lig-2019.csv",
"old_data/turkey-super-lig-2020.csv",
"old_data/epl-2009.csv",
"old_data/epl-2010.csv",
"old_data/epl-2011.csv",
"old_data/epl-2012.csv",
"old_data/epl-2013.csv",
"old_data/epl-2014.csv",
"old_data/epl-2015.csv",
"old_data/epl-2016.csv",
"old_data/epl-2017.csv",
"old_data/epl-2018.csv",
"old_data/epl-2019.csv",
"old_data/epl-2020.csv",
"epl-2021.csv",
],
getAllInputs=False,
starting_week=10):
total_inputs = []
total_outputs = []
for filename in filenames:
print("READING ", filename)
bothFuncs = [
addNewInputAndOutput,
addRankToInputs("A"),
addRankToInputs("A-week"),
addRankToInputs("A-G"),
addRankToInputs("A-G-week"),
createA("A", 1.0, 0.0, None),
createA("A-week", 1.0, 0.0, createWeekMultiplier(0.5)),
createA("A-G", 0.0, 1.0, None),
createA("A-G-week", 0.0, 1.0, createWeekMultiplier(0.5)),
addAverageToInputs(),
]
repeatFuncs = [
addNewInputAndOutput,
addRankToInputs("A"),
addRankToInputs("A-week"),
addRankToInputs("A-G"),
addRankToInputs("A-G-week"),
createA("A", 1.0, 0.0, None),
createA("A-week", 1.0, 0.0, createWeekMultiplier(0.5)),
createA("A-G", 0.0, 1.0, None),
createA("A-G-week", 0.0, 1.0, createWeekMultiplier(0.5)),
]
model = epl.getData(filename, False)
data = loopThroughData(model, bothFuncs, not (getAllInputs),
starting_week)
inputs = data["inputs"]
outputs = data["outputs"]
model = epl.getData(filename, True)
data = loopThroughData(model, repeatFuncs, not (getAllInputs),
starting_week)
inps = data["inputs"]
assert (len(inputs) == len(inps))
for i in range(len(inputs)):
inputs[i] += inps[i]
total_inputs += inputs
total_outputs += outputs
return np.array(total_inputs), np.array(total_outputs)
def predictGames(filename):
inputs, labels = cdata.getAllMyData([filename], True, 0)
data, indexToTeam, teamToIndex, indexToGamesPlayed = epl.getData(filename)
model = makeModel(inputs, inputs.shape[1], labels.shape[1])
outputs = model.predict(inputs)
print("SHAPE: ", inputs.shape, labels.shape, outputs.shape)
print(
"%20s : %s Goals %s %-20s Home Tie Away Accurate (Should be close to 1 for accuracy)"
% ("Home Team", " " * 2, " " * 5, "Away Team"))
indexToPoints = [0.0 for i in range(len(indexToTeam))]
for i in range(len(data["Home Team"])):
homeTeam = data["Home Team"][i]
awayTeam = data["Away Team"][i]
homeScore = None
awayScore = None
if (type(data["Result"][i]) is str):
result = data["Result"][i].split("-")
if len(result) == 2:
homeScore = int(result[0].strip())
awayScore = int(result[1].strip())
homeIndex = teamToIndex[homeTeam]
awayIndex = teamToIndex[awayTeam]
homeTeamPlayed = indexToGamesPlayed[homeIndex]
awayTeamPlayed = indexToGamesPlayed[awayIndex]
output = np.array(outputs[i])
p_output = output / output.sum()
if homeScore != None:
if homeScore > awayScore:
indexToPoints[homeIndex] += 3.0
elif homeScore < awayScore:
indexToPoints[awayIndex] += 3.0
else:
indexToPoints[homeIndex] += 1.0
indexToPoints[awayIndex] += 1.0
else:
indexToPoints[homeIndex] += p_output[0] * 3.0 + p_output[2]
indexToPoints[awayIndex] += p_output[1] * 3.0 + p_output[2]
# Print Tie
#print("%4.2f %4.2f %4.2f %4.2f" % (inputs[i][12], inputs[i][13], inputs[i][22], inputs[i][23]))
if homeScore == None:
print(
"%20s : %s vs %s : %-20s %4.2f %4.2f %4.2f %10.2f %8.2f"
% (homeTeam, " " * 4, " " * 4, awayTeam, p_output[0],
p_output[2], p_output[1], output.sum(), 1.0 -
(5 * (output.sum() - 1.0))**2))
#print("%20s : %s vs %s : %-20s %4.2f" %
# (homeTeam, " " * 4, " " * 4, awayTeam, output[0]))
else:
print(
"%20s : %4d vs %-4d : %-20s %4.2f %4.2f %4.2f %10.2f %8.2f"
% (homeTeam, homeScore, awayScore, awayTeam, p_output[0],
p_output[2], p_output[1], output.sum(), 1.0 -
(5 * (output.sum() - 1.0))**2))
#print("%20s : %4d vs %-4d : %-20s %4.2f" %
# (homeTeam, homeScore, awayScore, awayTeam, output[0]))
points = np.array([indexToPoints, np.arange(len(indexToPoints))])
points = np.array(sorted(points.T, key=lambda x: x[0], reverse=True))
print("\n\n")
print(" %-30s %5s" % ("Team", "Points"))
for i in range(len(points)):
print("%2d %-30s %5.3f" %
(i + 1, indexToTeam[int(points[i][1])], points[i][0]))
def calculateProbs(predictionCount, csv_file_name, predict):
data, indexToTeam, teamToIndex, indexToGamesPlayed = epl.getData(
csv_file_name)
indexToPlaceFinishedToTimesFinished = [
[0 for _ in range(len(indexToTeam))] for _ in range(len(indexToTeam))
]
indexToPoints = [0 for _ in range(len(indexToTeam))]
indexToExpectedPoints = np.array([0.0 for _ in range(len(indexToTeam))])
gameIndexLeftToPlay = []
for i in range(len(data["Home Team"])):
homeTeam = data["Home Team"][i]
awayTeam = data["Away Team"][i]
homeIndex = teamToIndex[homeTeam]
awayIndex = teamToIndex[awayTeam]
if not (type(data["Result"][i]) is str):
gameIndexLeftToPlay.append(i)
winnerProp, tieProb, loserProb = predict(homeIndex, awayIndex, i)
total = winnerProp + tieProb + loserProb
winnerProp = winnerProp / total
tieProb = tieProb / total
loserProb = loserProb / total
indexToExpectedPoints[homeIndex] += winnerProp * 3 + tieProb
indexToExpectedPoints[awayIndex] += loserProb * 3 + tieProb
continue
result = data["Result"][i].split("-")
if len(result) != 2:
gameIndexLeftToPlay.append(i)
winnerProp, tieProb, loserProb = predict(homeIndex, awayIndex, i)
total = winnerProp + tieProb + loserProb
winnerProp = winnerProp / total
tieProb = tieProb / total
loserProb = loserProb / total
indexToExpectedPoints[homeIndex] += winnerProp * 3 + tieProb
indexToExpectedPoints[awayIndex] += loserProb * 3 + tieProb
continue
homeScore = int(result[0].strip())
awayScore = int(result[1].strip())
if homeScore > awayScore:
indexToPoints[homeIndex] += 3
indexToExpectedPoints[homeIndex] += 3
elif homeScore < awayScore:
indexToPoints[awayIndex] += 3
indexToExpectedPoints[awayIndex] += 3
else:
indexToPoints[homeIndex] += 1
indexToPoints[awayIndex] += 1
indexToExpectedPoints[homeIndex] += 1
indexToExpectedPoints[awayIndex] += 1
for rounds in range(predictionCount):
if rounds % 1000 == 0:
print(rounds, (rounds / predictionCount))
roundIndexToPoints = indexToPoints.copy()
for i in gameIndexLeftToPlay:
homeTeam = data["Home Team"][i]
awayTeam = data["Away Team"][i]
homeIndex = teamToIndex[homeTeam]
awayIndex = teamToIndex[awayTeam]
winnerProp, tieProb, loserProb = predict(homeIndex, awayIndex, i)
total = winnerProp + tieProb + loserProb
winnerProp = winnerProp / total
tieProb = tieProb / total
if random.random() <= winnerProp:
roundIndexToPoints[homeIndex] += 3
elif random.random() <= winnerProp + tieProb:
roundIndexToPoints[homeIndex] += 1
roundIndexToPoints[homeIndex] += 1
else:
roundIndexToPoints[awayIndex] += 3
rankings = [(roundIndexToPoints[i], i)
for i in range(len(roundIndexToPoints))]
rankings.sort(reverse=True)
for i in range(len(rankings)):
indexToPlaceFinishedToTimesFinished[rankings[i][1]][i] += 1
return np.array(indexToPlaceFinishedToTimesFinished
) / predictionCount, indexToExpectedPoints, indexToTeam