def pred(facility, model, resolution):
filename, columns, irrelevantColumns, targetColumns, traintime, testtime, columnOrder = configs.getConfig(
facility, model, resolution)
df = mlModule.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlModule.getTestTrainSplit(traintime, testtime)
X_train, y_train, X_test, y_test = mlModule.getFeatureTargetSplit(
targetColumns)
lstm_1_1 = mlModule.LSTM('LSTM 1x128 d0.0' + ' mod' + model,
layers=[128],
dropout=0.0,
recurrentDropout=0.0,
epochs=5000)
lstm_1_2 = mlModule.LSTM('LSTM 1x128 d0.1' + ' mod' + model,
layers=[128],
dropout=0.1,
recurrentDropout=0.1,
epochs=5000)
lstm_1_3 = mlModule.LSTM('LSTM 1x128 d0.2' + ' mod' + model,
layers=[128],
dropout=0.2,
recurrentDropout=0.2,
epochs=5000)
lstm_1_4 = mlModule.LSTM('LSTM 1x128 d0.3' + ' mod' + model,
layers=[128],
dropout=0.3,
recurrentDropout=0.3,
epochs=5000)
lstm_1_5 = mlModule.LSTM('LSTM 1x128 d0.4' + ' mod' + model,
layers=[128],
dropout=0.4,
recurrentDropout=0.4,
epochs=5000)
lstm_1_6 = mlModule.LSTM('LSTM 1x128 d0.5' + ' mod' + model,
layers=[128],
dropout=0.5,
recurrentDropout=0.5,
epochs=5000)
linear = mlModule.Linear_Regularized('Linear rCV mod' + model)
initTrainPredict([
linear,
lstm_1_1,
lstm_1_2,
lstm_1_3,
lstm_1_4,
lstm_1_5,
lstm_1_6,
])
def pred(facility, model, resolution):
filename, columns, irrelevantColumns, targetColumns, traintime, testtime, columnOrder = configs.getConfig(facility, model, resolution)
df = mlModule.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlModule.getTestTrainSplit(traintime, testtime)
X_train, y_train, X_test, y_test = mlModule.getFeatureTargetSplit(targetColumns)
mlp_mae = mlModule.MLP('MLP 1x128 d0.2 mae mod'+model, layers=[128], dropout=0.2, loss='mean_absolute_error', metrics=['mean_absolute_error'])
mlp_mse = mlModule.MLP('MLP 1x128 d0.2 mse mod'+model, layers=[128], dropout=0.2, loss='mean_squared_error', metrics=['mean_squared_error'])
lstm_mae = mlModule.LSTM('LSTM 1x128 d0.2 mae mod'+model, layers=[128], dropout=0.2, recurrentDropout=0.2, enrolWindow=12, loss='mean_absolute_error', metrics=['mean_absolute_error'])
lstm_mse = mlModule.LSTM('LSTM 1x128 d0.2 mse mod'+model, layers=[128], dropout=0.2, recurrentDropout=0.2, enrolWindow=12, loss='mean_squared_error', metrics=['mean_squared_error'])
modelList = [
mlp_mae,
mlp_mse,
lstm_mae,
lstm_mse,
]
initTrainPredict(modelList)
def performDropoutPrediction(facility,
model,
resolution,
lookback=12,
retrain=False):
filename, columns, irrelevantColumns, targetColumns, traintime, testtime, columnOrder = configs.getConfig(
facility, model, resolution)
df = mlModule.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlModule.getTestTrainSplit(traintime, testtime)
X_train, y_train, X_test, y_test = mlModule.getFeatureTargetSplit(
targetColumns)
lstm = mlModule.LSTM('LSTMs 1x128 d0.2 mod' + model,
layers=[128],
training=True,
dropout=0.2,
recurrentDropout=0.2,
enrolWindow=lookback)
gru = mlModule.GRU('GRUs 1x128 d0.2 mod' + model,
layers=[128],
training=True,
dropout=0.2,
recurrentDropout=0.2,
enrolWindow=lookback)
modelList = [
lstm,
gru,
]
mlModule.initModels(modelList)
mlModule.trainModels(retrain)
predictions, means, stds = mlModule.predictWithModelsUsingDropout(
numberOfPredictions=30)
plotDropoutPrediction(modelList, predictions, means, stds, targetColumns,
df_test, y_test, traintime)
# List of column names used a targets
targetColumns = [
'50TT002',
'20PDT001',
]
# List of training periods on form ['start', 'end']
traintime = [
["2020-01-01 00:00:00", "2020-03-20 00:00:00"],
]
# Testing period, recommended: entire dataset
testtime = ["2020-01-01 00:00:00", "2020-08-01 00:00:00"]
df = mlModule.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlModule.getTestTrainSplit(traintime, testtime)
X_train, y_train, X_test, y_test = mlModule.getFeatureTargetSplit(
targetColumns)
mlp_1 = mlModule.MLP('MLP 1x64 d0.2 mod' + model, layers=[64], dropout=0.2)
mlp_2 = mlModule.MLP('MLP 1x128 d0.2 mod' + model, layers=[128], dropout=0.2)
mlp_3 = mlModule.MLP('MLP 2x64 d0.2 mod' + model, layers=[64, 64], dropout=0.2)
mlp_4 = mlModule.MLP('MLP 2x128 d0.2 mod' + model,
layers=[128, 128],
dropout=0.2)
lstm_1 = mlModule.LSTM('LSTM 1x64 d0.2 mod' + model,
layers=[64],
dropout=0.2,
recurrentDropout=0.2,
enrolWindow=12)
def featureComparison(
irrelevantColumnsList,
filename,
columns,
traintime,
testtime,
targetColumns,
enrolWindow,
):
global colors, models
columnsLists = []
deviationsLists = []
names = []
trainmetrics = []
testmetrics = []
for i, irrelevantColumns in enumerate(irrelevantColumnsList):
mlModule.reset()
df = mlModule.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlModule.getTestTrainSplit(traintime, testtime)
X_train, y_train, X_test, y_test = mlModule.getFeatureTargetSplit(
targetColumns)
mlp_1 = mlModule.MLP('MLP 1x64 d0.2 mod' + models[i],
layers=[64],
dropout=0.2)
mlp_2 = mlModule.MLP('MLP 1x128 d0.2 mod' + models[i],
layers=[128],
dropout=0.2)
mlp_3 = mlModule.MLP('MLP 2x64 d0.2 mod' + models[i],
layers=[64, 64],
dropout=0.2)
mlp_4 = mlModule.MLP('MLP 2x128 d0.2 mod' + models[i],
layers=[128, 128],
dropout=0.2)
lstm_1 = mlModule.LSTM('LSTM 1x64 d0.2 mod' + models[i],
layers=[64],
dropout=0.2,
recurrentDropout=0.2,
enrolWindow=12)
lstm_2 = mlModule.LSTM('LSTM 1x128 d0.2 mod' + models[i],
layers=[128],
dropout=0.2,
recurrentDropout=0.2,
enrolWindow=12)
lstm_3 = mlModule.LSTM('LSTM 2x64 d0.2 mod' + models[i],
layers=[64, 64],
dropout=0.2,
recurrentDropout=0.2,
enrolWindow=12)
lstm_4 = mlModule.LSTM('LSTM 2x128 d0.2 mod' + models[i],
layers=[128, 128],
dropout=0.2,
recurrentDropout=0.2,
enrolWindow=12)
linear = mlModule.Linear_Regularized('Linear rCV mod' + models[i])
modelList = [
mlp_1,
mlp_2,
mlp_3,
mlp_4,
lstm_1,
lstm_2,
lstm_3,
lstm_4,
linear,
]
mlModule.initModels(modelList)
retrain = False
mlModule.trainModels(retrain)
modelNames, metrics_train, metrics_test, columnsList, deviationsList = mlModule.predictWithModels(
plot=True, score=True)
if i < 1:
columnsLists = columnsList
deviationsLists = deviationsList
all_names = modelNames
all_train_metrics = metrics_train
all_test_metrics = metrics_test
else:
for j_target in range(len(columnsList)):
for k_model in range(1, len(columnsList[j_target])):
columnsLists[j_target].append(
columnsList[j_target][k_model])
for k_model in range(0, len(deviationsList[j_target])):
deviationsLists[j_target].append(
deviationsList[j_target][k_model])
all_names = [*all_names, *modelNames]
all_train_metrics = [*all_train_metrics, *metrics_train]
all_test_metrics = [*all_test_metrics, *metrics_test]
names.append(modelNames)
trainmetrics.append(metrics_train)
testmetrics.append(metrics_test)
indexColumn = mlModule._indexColumn
columnDescriptions = mlModule._columnDescriptions
columnUnits = mlModule._columnUnits
traintime = mlModule._traintime
for i in range(len(deviationsLists)):
for j in range(len(deviationsLists[i])):
deviationsLists[i][j][3] = colors[j]
for i in range(len(columnsLists)):
columnsList[i][0][3] = 'red'
for j in range(1, len(columnsLists[i])):
columnsLists[i][j][3] = colors[j - 1]
printModelScores(
all_names,
all_train_metrics,
all_test_metrics,
)
plotModelPredictions(
plt,
deviationsLists,
columnsLists,
indexColumn,
columnDescriptions,
columnUnits,
traintime,
interpol=False,
)
plotModelScores(
plt,
all_names,
all_train_metrics,
all_test_metrics,
)
def pred(facility, model, resolution):
filename, columns, irrelevantColumns, targetColumns, traintime, testtime, columnOrder = configs.getConfig(
facility, model, resolution)
df = mlModule.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlModule.getTestTrainSplit(traintime, testtime)
X_train, y_train, X_test, y_test = mlModule.getFeatureTargetSplit(
targetColumns)
mlpr_1_1 = mlModule.MLP('MLP 1x128 1.0' + ' mod' + model,
layers=[128],
l1_rate=1.0,
epochs=5000)
mlpr_1_2 = mlModule.MLP('MLP 1x128 0.5' + ' mod' + model,
layers=[128],
l1_rate=0.5,
epochs=5000)
mlpr_1_3 = mlModule.MLP('MLP 1x128 0.1' + ' mod' + model,
layers=[128],
l1_rate=0.1,
epochs=5000)
mlpr_1_4 = mlModule.MLP('MLP 1x128 0.05' + ' mod' + model,
layers=[128],
l1_rate=0.05,
epochs=5000)
mlpr_1_5 = mlModule.MLP('MLP 1x128 0.01' + ' mod' + model,
layers=[128],
l1_rate=0.01,
epochs=5000)
mlpr_1_6 = mlModule.MLP('MLP 1x128 0.005' + ' mod' + model,
layers=[128],
l1_rate=0.005,
epochs=5000)
mlpr_1_7 = mlModule.MLP('MLP 1x128 0.001' + ' mod' + model,
layers=[128],
l1_rate=0.001,
epochs=5000)
mlpd_1_8 = mlModule.MLP('MLP 1x128 0.2' + ' mod' + model,
layers=[128],
dropout=0.2,
epochs=5000)
linear_r = mlModule.Linear_Regularized('Linear rCV' + ' mod' + model)
initTrainPredict([
mlpr_1_1,
mlpr_1_2,
mlpr_1_3,
mlpr_1_4,
mlpr_1_5,
mlpr_1_6,
mlpr_1_7,
mlpd_1_8,
linear_r,
])
initTrainPredict([
mlpr_1_6,
mlpr_1_7,
mlpd_1_8,
linear_r,
])
["2020-01-01 00:00:00", "2020-04-01 00:00:00"],
]
# In this case, two separate testing phases are used
# This testtime parameter should cover the entire dataset
testtime = ["2020-01-01 00:00:00", "2020-08-01 00:00:00"]
# This testtime1 parameter should cover the first testing phase
testtime1 = ["2020-04-15 00:00:00", "2020-05-04 00:00:00"]
# This testtime2 parameter should cover the second testing phase
testtime2 = ["2020-06-01 00:00:00", "2020-06-16 00:00:00"]
# 2. Initiate and divide data
df = mlApi.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlApi.getTestTrainSplit(traintime, testtime)
df_test_1, df_test_2 = mlApi.getTestTrainSplit([testtime1], testtime2)
df_test_joined = pd.concat([df_test_1, df_test_2])
# 3. Plot correlation plots
mlApi.correlationPlot(df_train, datasetName + " train")
mlApi.correlationDuoPlot(df_test_1, df_test_2, datasetName + " test 1",
datasetName + " test 2")
mlApi.correlationDifferencePlot(
df_train, df_test_joined, "Difference, " + datasetName + " train and test")
# Reset to prepare for second dataset
# -------------------------------------
mlApi.reset()
def pred(facility, model, resolution):
filename, columns, irrelevantColumns, targetColumns, traintime, testtime, columnOrder = configs.getConfig(
facility, model, resolution)
df = mlModule.initDataframe(filename, columns, irrelevantColumns)
df_train, df_test = mlModule.getTestTrainSplit(traintime, testtime)
X_train, y_train, X_test, y_test = mlModule.getFeatureTargetSplit(
targetColumns)
mlp_1x_16 = mlModule.MLP('MLP 1x16' + ' mod' + model,
layers=[16],
dropout=0.2,
epochs=1000)
mlp_1x_32 = mlModule.MLP('MLP 1x32' + ' mod' + model,
layers=[32],
dropout=0.2,
epochs=1000)
mlp_1x_64 = mlModule.MLP('MLP 1x64' + ' mod' + model,
layers=[64],
dropout=0.2,
epochs=1000)
mlp_1x_128 = mlModule.MLP('MLP 1x128' + ' mod' + model,
layers=[128],
dropout=0.2,
epochs=1000)
mlp_2x_16 = mlModule.MLP('MLP 2x16' + ' mod' + model,
layers=[16, 16],
dropout=0.2,
epochs=1000)
mlp_2x_32 = mlModule.MLP('MLP 2x32' + ' mod' + model,
layers=[32, 32],
dropout=0.2,
epochs=1000)
mlp_2x_64 = mlModule.MLP('MLP 2x64' + ' mod' + model,
layers=[64, 64],
dropout=0.2,
epochs=1000)
mlp_2x_128 = mlModule.MLP('MLP 2x128' + ' mod' + model,
layers=[128, 128],
dropout=0.2,
epochs=1000)
linear_cv = mlModule.Linear_Regularized('Linear rCV' + ' mod' + model)
ensemble = mlModule.Ensemble('MLP 1x128 + Linear' + ' mod' + model,
[mlp_1x_128, linear_cv])
ensemble2 = mlModule.Ensemble('MLP 2x64 + Linear' + ' mod' + model,
[mlp_2x_64, linear_cv])
modelList = [
linear_cv,
mlp_1x_16,
mlp_1x_32,
mlp_2x_16,
mlp_2x_32,
]
initTrainPredict(modelList)
modelList = [
linear_cv,
mlp_1x_64,
mlp_1x_128,
mlp_2x_64,
mlp_2x_128,
]
initTrainPredict(modelList)
modelList = [
linear_cv,
ensemble,
ensemble2,
]
initTrainPredict(modelList)