import thesisFunctions months = ['jan', 'feb', 'mar', 'apr', 'may', 'jun', 'jul', 'aug', 'sep', 'oct', 'nov', 'dec'] regions = ['IntMnt', 'Xeric'] dryFolder = 'AllMonthsDryHalf/' wetFolder = 'AllMonthsWetHalf/' baseFolders = [dryFolder, wetFolder] outputFolder = 'Output/' outputFileName = 'AllMonthsData.csv' # Aggregate each half-region-month's predictions into one file. thesisFunctions.aggregateSacPredictions(baseFolders, outputFolder, outputFileName, months, regions) # Output IntMnt file for specific water year for use in DWRAT aggregateFile = outputFolder + outputFileName waterYear = 1977 thesisFunctions.formatWaterYearPredictions(waterYear, aggregateFile)
import thesisFunctions
months = [
'jan', 'feb', 'mar', 'apr', 'may', 'jun', 'jul', 'aug', 'sep', 'oct',
'nov', 'dec'
]
regions = ['IntMnt', 'Xeric']
dryFolder = 'AllMonthsDryHalf/'
wetFolder = 'AllMonthsWetHalf/'
baseFolders = [dryFolder, wetFolder]
outputFolder = 'Output/'
outputFileName = 'AllMonthsData.csv'
# Aggregate each half-region-month's predictions into one file.
thesisFunctions.aggregateSacPredictions(baseFolders, outputFolder,
outputFileName, months, regions)
# Output IntMnt file for specific water year for use in DWRAT
aggregateFile = outputFolder + outputFileName
waterYear = 1977
thesisFunctions.formatWaterYearPredictions(waterYear, aggregateFile)
def runModels(basePath, performanceEstimation=True, prediction=False):
randomSeed = constants.randomSeed
myFeaturesIndex = 6
myLabelIndex = 5
kFolds = 5
regions = ['IntMnt', 'Xeric']
months = ['jan', 'feb', 'mar', 'apr', 'may', 'jun', 'jul', 'aug', 'sep', 'oct', 'nov', 'dec']
r2Method = mltypes.ModelScoreMethod('R Squared', sklearn.metrics.r2_score)
meanOEMethod = mltypes.ModelScoreMethod('Mean O/E', mlmodel.meanObservedExpectedScore)
sdOEMethod = mltypes.ModelScoreMethod('Standard Deviation O/E', mlmodel.sdObservedExpectedScore)
mseMethod = mltypes.ModelScoreMethod('Mean Squared Error (cfs)', sklearn.metrics.mean_squared_error)
testScoreMethods = [r2Method, meanOEMethod, sdOEMethod, mseMethod]
randomForestParameters = {'n_estimators': 2000,
'max_features': .333,
'random_state': randomSeed,
'n_jobs': -1}
randomForestMethod = mltypes.ModellingMethod(constants.randomForest,
sklearn.ensemble.RandomForestRegressor)
for region in regions:
for month in months:
print('Processing:', region, month.capitalize())
# Get expert features from text files
selectedFeatures = getMonthVars(basePath, month, region)
expertSelectedConfig = mltypes.FeatureEngineeringConfiguration('Expert Selection',
'selection',
mltypes.ExtractSpecificFeatures,
{'featureList': selectedFeatures})
modelFolder = basePath + region + '/' + month + '/'
# Run model once on each fold to get estimates of test metrics
if performanceEstimation:
allFoldScoreModelResultsDFs = []
for fold in range(kFolds):
# Get dataset info
foldTestFilePath = modelFolder + '{}_{}_{}_test.csv'.format(month, region, fold)
foldTrainFilePath = modelFolder + '{}_{}_all_{}_train.csv'.format(month, region, fold)
testDescription = month.capitalize() + ' ' + region + ' Test'
trainDescription = month.capitalize() + ' ' + region + ' Train'
# Copy to CurrentFoldDataFolder
testFilePath = modelFolder + 'CurrentFoldData/' + '{}_{}_test.csv'.format(month, region)
trainFilePath = modelFolder + 'CurrentFoldData/' + '{}_{}_all_train.csv'.format(month, region)
shutil.copyfile(foldTestFilePath, testFilePath)
shutil.copyfile(foldTrainFilePath, trainFilePath)
# Get datasets
fullTestDataSet = mltypes.DataSet(testDescription,
testFilePath,
featuresIndex=myFeaturesIndex,
labelIndex=myLabelIndex)
fullTrainDataSet = mltypes.DataSet(trainDescription,
trainFilePath,
featuresIndex=myFeaturesIndex,
labelIndex=myLabelIndex)
fullTrainDataSet = makeLabelRunoffPerDrainageUnit(fullTrainDataSet, 'labeled')
fullTestDataSet = makeLabelRunoffPerDrainageUnit(fullTestDataSet, 'labeled')
# Select features
trainDataSet, transformer = mldata.engineerFeaturesForDataSet(fullTrainDataSet,
expertSelectedConfig)
testDataSet = mldata.engineerFeaturesByTransformer(fullTestDataSet,
transformer)
# Apply model
applyRFModelConfig = mltypes.ApplyModelConfiguration('Apply ' + constants.randomForest,
randomForestMethod,
randomForestParameters,
trainDataSet,
testDataSet)
randomForestResult = mlmodel.applyModel(applyRFModelConfig)
applyModelResults = [randomForestResult]
# Score model and convert results to data frame
scoreModelResults = mlmodel.scoreModels(applyModelResults, testScoreMethods)
scoreModelResultsDF = mlutils.createScoreDataFrame(scoreModelResults)
# Add RMSE, then add to list of results for this month
scoreModelResultsDF['RMSE (cfs)'] = scoreModelResultsDF['Mean Squared Error (cfs)'].map(lambda x: x ** (1/2))
allFoldScoreModelResultsDFs.append(scoreModelResultsDF)
print(region, month, fold, 'processed')
# Aggregate results into a single DataFrame
allResultsDF = pandas.DataFrame()
for fold in allFoldScoreModelResultsDFs:
allResultsDF = allResultsDF.append(fold, ignore_index=True)
allResultsDF.to_csv(modelFolder + 'Output/scoreModelResults_all.csv', index=False)
# Group by unique model & dataset combinations to average
averageResultsDF = allResultsDF.groupby(['Base DataSet', 'Model Method']).mean().reset_index()
sortedAverageResultsDF = averageResultsDF.sort(columns='R Squared', ascending=False)
sortedAverageResultsDF.to_csv(modelFolder + 'Output/scoreModelResults_average.csv', index=False)
# Prediction
if prediction:
predictionFolder = modelFolder + 'Prediction/'
# Get data
fullTrainDataSet = mltypes.DataSet(month.capitalize() + ' Training Data',
predictionFolder + '{}_{}_all.csv'.format(month, region),
featuresIndex=myFeaturesIndex,
labelIndex=myLabelIndex)
fullPredictionDataSet = mltypes.DataSet(month.capitalize() + ' Prediction Data',
predictionFolder + 'sacramentoData.csv',
featuresIndex=3,
labelIndex=None)
# Get scaled label (runoff/drainage unit)
fullTrainDataSet = makeLabelRunoffPerDrainageUnit(fullTrainDataSet, 'labeled')
fullPredictionDataSet = makeLabelRunoffPerDrainageUnit(fullPredictionDataSet, 'prediction')
# Select features
trainDataSet, transformer = mldata.engineerFeaturesForDataSet(fullTrainDataSet,
expertSelectedConfig)
predictionDataSet = mldata.engineerFeaturesByTransformer(fullPredictionDataSet,
transformer)
# Train model and predict for the Sacramento region
applyRFModelConfig = mltypes.ApplyModelConfiguration('Apply ' + constants.randomForest,
randomForestMethod,
randomForestParameters,
trainDataSet,
predictionDataSet)
applyRFModelResult = mlmodel.applyModel(applyRFModelConfig)
rescalePredictions(applyRFModelResult, predictionDataSet)
predictionOutputPath = predictionFolder + 'sacramentoPredictions.csv'
thesisFunctions.outputPredictions(applyRFModelResult, predictionOutputPath)
if prediction:
print('Aggregating predictions.')
aggregateFile = thesisFunctions.aggregateSacPredictions([basePath],
'Output/',
'RandomForestData.csv',
months,
regions)
waterYear = 1977
thesisFunctions.formatWaterYearPredictions(waterYear, aggregateFile)
def runModels(basePath, performanceEstimation=True, prediction=False):
randomSeed = constants.randomSeed
myFeaturesIndex = 6
myLabelIndex = 5
kFolds = 5
regions = ['IntMnt', 'Xeric']
months = [
'jan', 'feb', 'mar', 'apr', 'may', 'jun', 'jul', 'aug', 'sep', 'oct',
'nov', 'dec'
]
r2Method = mltypes.ModelScoreMethod('R Squared', sklearn.metrics.r2_score)
meanOEMethod = mltypes.ModelScoreMethod('Mean O/E',
mlmodel.meanObservedExpectedScore)
sdOEMethod = mltypes.ModelScoreMethod('Standard Deviation O/E',
mlmodel.sdObservedExpectedScore)
mseMethod = mltypes.ModelScoreMethod('Mean Squared Error (cfs)',
sklearn.metrics.mean_squared_error)
testScoreMethods = [r2Method, meanOEMethod, sdOEMethod, mseMethod]
randomForestParameters = {
'n_estimators': 2000,
'max_features': .333,
'random_state': randomSeed,
'n_jobs': -1
}
randomForestMethod = mltypes.ModellingMethod(
constants.randomForest, sklearn.ensemble.RandomForestRegressor)
for region in regions:
for month in months:
print('Processing:', region, month.capitalize())
# Get expert features from text files
selectedFeatures = getMonthVars(basePath, month, region)
expertSelectedConfig = mltypes.FeatureEngineeringConfiguration(
'Expert Selection', 'selection',
mltypes.ExtractSpecificFeatures,
{'featureList': selectedFeatures})
modelFolder = basePath + region + '/' + month + '/'
# Run model once on each fold to get estimates of test metrics
if performanceEstimation:
allFoldScoreModelResultsDFs = []
for fold in range(kFolds):
# Get dataset info
foldTestFilePath = modelFolder + '{}_{}_{}_test.csv'.format(
month, region, fold)
foldTrainFilePath = modelFolder + '{}_{}_all_{}_train.csv'.format(
month, region, fold)
testDescription = month.capitalize(
) + ' ' + region + ' Test'
trainDescription = month.capitalize(
) + ' ' + region + ' Train'
# Copy to CurrentFoldDataFolder
testFilePath = modelFolder + 'CurrentFoldData/' + '{}_{}_test.csv'.format(
month, region)
trainFilePath = modelFolder + 'CurrentFoldData/' + '{}_{}_all_train.csv'.format(
month, region)
shutil.copyfile(foldTestFilePath, testFilePath)
shutil.copyfile(foldTrainFilePath, trainFilePath)
# Get datasets
fullTestDataSet = mltypes.DataSet(
testDescription,
testFilePath,
featuresIndex=myFeaturesIndex,
labelIndex=myLabelIndex)
fullTrainDataSet = mltypes.DataSet(
trainDescription,
trainFilePath,
featuresIndex=myFeaturesIndex,
labelIndex=myLabelIndex)
fullTrainDataSet = makeLabelRunoffPerDrainageUnit(
fullTrainDataSet, 'labeled')
fullTestDataSet = makeLabelRunoffPerDrainageUnit(
fullTestDataSet, 'labeled')
# Select features
trainDataSet, transformer = mldata.engineerFeaturesForDataSet(
fullTrainDataSet, expertSelectedConfig)
testDataSet = mldata.engineerFeaturesByTransformer(
fullTestDataSet, transformer)
# Apply model
applyRFModelConfig = mltypes.ApplyModelConfiguration(
'Apply ' + constants.randomForest, randomForestMethod,
randomForestParameters, trainDataSet, testDataSet)
randomForestResult = mlmodel.applyModel(applyRFModelConfig)
applyModelResults = [randomForestResult]
# Score model and convert results to data frame
scoreModelResults = mlmodel.scoreModels(
applyModelResults, testScoreMethods)
scoreModelResultsDF = mlutils.createScoreDataFrame(
scoreModelResults)
# Add RMSE, then add to list of results for this month
scoreModelResultsDF['RMSE (cfs)'] = scoreModelResultsDF[
'Mean Squared Error (cfs)'].map(lambda x: x**(1 / 2))
allFoldScoreModelResultsDFs.append(scoreModelResultsDF)
print(region, month, fold, 'processed')
# Aggregate results into a single DataFrame
allResultsDF = pandas.DataFrame()
for fold in allFoldScoreModelResultsDFs:
allResultsDF = allResultsDF.append(fold, ignore_index=True)
allResultsDF.to_csv(modelFolder +
'Output/scoreModelResults_all.csv',
index=False)
# Group by unique model & dataset combinations to average
averageResultsDF = allResultsDF.groupby(
['Base DataSet', 'Model Method']).mean().reset_index()
sortedAverageResultsDF = averageResultsDF.sort(
columns='R Squared', ascending=False)
sortedAverageResultsDF.to_csv(
modelFolder + 'Output/scoreModelResults_average.csv',
index=False)
# Prediction
if prediction:
predictionFolder = modelFolder + 'Prediction/'
# Get data
fullTrainDataSet = mltypes.DataSet(
month.capitalize() + ' Training Data',
predictionFolder + '{}_{}_all.csv'.format(month, region),
featuresIndex=myFeaturesIndex,
labelIndex=myLabelIndex)
fullPredictionDataSet = mltypes.DataSet(
month.capitalize() + ' Prediction Data',
predictionFolder + 'sacramentoData.csv',
featuresIndex=3,
labelIndex=None)
# Get scaled label (runoff/drainage unit)
fullTrainDataSet = makeLabelRunoffPerDrainageUnit(
fullTrainDataSet, 'labeled')
fullPredictionDataSet = makeLabelRunoffPerDrainageUnit(
fullPredictionDataSet, 'prediction')
# Select features
trainDataSet, transformer = mldata.engineerFeaturesForDataSet(
fullTrainDataSet, expertSelectedConfig)
predictionDataSet = mldata.engineerFeaturesByTransformer(
fullPredictionDataSet, transformer)
# Train model and predict for the Sacramento region
applyRFModelConfig = mltypes.ApplyModelConfiguration(
'Apply ' + constants.randomForest, randomForestMethod,
randomForestParameters, trainDataSet, predictionDataSet)
applyRFModelResult = mlmodel.applyModel(applyRFModelConfig)
rescalePredictions(applyRFModelResult, predictionDataSet)
predictionOutputPath = predictionFolder + 'sacramentoPredictions.csv'
thesisFunctions.outputPredictions(applyRFModelResult,
predictionOutputPath)
if prediction:
print('Aggregating predictions.')
aggregateFile = thesisFunctions.aggregateSacPredictions(
[basePath], 'Output/', 'RandomForestData.csv', months, regions)
waterYear = 1977
thesisFunctions.formatWaterYearPredictions(waterYear, aggregateFile)