def evalModel(self, X, numTrain):
"""Evaluate model from training"""
pdf = pd.DataFrame()
evaluator = RegressionEvaluator(metricName="rmse", labelCol="rating", predictionCol="prediction")
count = 0
for i in range(numTrain):
(train, test) = X.randomSplit([0.8, 0.2])
model = ALS(implicitPrefs=True,
rank=self.bestRank,
maxIter=self.bestNumIter,
regParam=self.bestLambda,
alpha=self.bestAlpha,
userCol="steamid",
itemCol="appid", ratingCol="rating").fit(train)
predictions = model.transform(test)
ones = predictions.where("rating=1")
zeroes = predictions.where("rating=0")
predictors = {'all': predictions, 'zeroes': zeroes, 'ones': ones}
for s, p in predictors.items():
pdf = pdf.append(pd.DataFrame([[i, s,
evaluator.setParams(metricName="rmse").evaluate(p),
evaluator.setParams(metricName="mse").evaluate(p),
evaluator.setParams(metricName="mae").evaluate(p)]]))
count += 1
print(round((i / 10) * 100, 0), '%')
pdf.columns = ['iteration', 'type', 'rmse', 'mse', 'mae']
print(pdf)
print(pdf.groupby(by=['type'], axis=0).mean())
bestNumIter = numIter
bestAlpha = alf
count += 1
#print('\r{0}%'.format(round((count / len(indexes)) * 100, 0)), end="", flush=True)
print("The best model was trained on evalData with rank = %d, lambda = %.2f, alpha = %d, " % (bestRank, bestLambda, bestAlpha) \
+ "numIter = %d and RMSE %f." % (bestNumIter, bestValidationRmse))
# brier score
# AUC
targetPrediction = bestModel.transform(target)
print('target prediction', targetPrediction.collect())
print('target RMSE:', evaluator.setParams(metricName="rmse").evaluate(targetPrediction))
print('target MSE:', evaluator.setParams(metricName="mse").evaluate(targetPrediction))
print('target MAE:', evaluator.setParams(metricName="mae").evaluate(targetPrediction))
predictions = bestModel.transform(test)
setvalues = ['all', 'zeroes', 'ones']
em = pd.DataFrame(columns=['rmse', 'mse', 'mae'])
em.index.names = ["set values"]
ones = predictions.where("rating=1")
zeroes = predictions.where("rating=0")
predictors = {'all': predictions, 'zeroes': zeroes, 'ones': ones}
#fpr, tpr, thresholds = roc_curve(predictions, pred, pos_label=2)
bestNumIter = numIter
bestAlpha = alf
count += 1
#print('\r{0}%'.format(round((count / len(indexes)) * 100, 0)), end="", flush=True)
print("The best model was trained on evalData with rank = %d, lambda = %.2f, alpha = %d, " % (bestRank, bestLambda, bestAlpha) \
+ "numIter = %d and RMSE %f." % (bestNumIter, bestValidationRmse))
# brier score
# AUC
targetPrediction = bestModel.transform(target)
print('target prediction', targetPrediction.collect())
print('target RMSE:',
evaluator.setParams(metricName="rmse").evaluate(targetPrediction))
print('target MSE:',
evaluator.setParams(metricName="mse").evaluate(targetPrediction))
print('target MAE:',
evaluator.setParams(metricName="mae").evaluate(targetPrediction))
predictions = bestModel.transform(test)
setvalues = ['all', 'zeroes', 'ones']
em = pd.DataFrame(columns=['rmse', 'mse', 'mae'])
em.index.names = ["set values"]
ones = predictions.where("rating=1")
zeroes = predictions.where("rating=0")
predictors = {'all': predictions, 'zeroes': zeroes, 'ones': ones}
regParam=bestLambda,
maxIter=bestNumIter,
alpha=bestAlpha,
userCol="steamid",
itemCol="appid",
ratingCol="rating").fit(train)
predictions = model.transform(test)
ones = predictions.where("rating=1")
zeroes = predictions.where("rating=0")
predictors = {'all': predictions, 'zeroes': zeroes, 'ones': ones}
for s, p in predictors.items():
pdf = pdf.append(
pd.DataFrame([[
i, s,
evaluator.setParams(metricName="rmse").evaluate(p),
evaluator.setParams(metricName="mse").evaluate(p),
evaluator.setParams(metricName="mae").evaluate(p)
]]))
count += 1
print(round((i / 10) * 100, 0), '%')
pdf.columns = ['iteration', 'type', 'rmse', 'mse', 'mae']
print(pdf)
print(pdf.groupby(by=['type'], axis=0).mean())
# brier score
# AUC
# setvalues = ['all', 'zeroes', 'ones']
#
# em = pd.DataFrame(columns=['rmse', 'mse', 'mae'])
# em.index.names = ["set values"]
