def outputProbabilisticForecastingResults(probabilisticForecastingResults,
savePath):
if probabilisticForecastingResults is not None:
pdfs = probabilisticForecastingResults["pdfs"]
cdfs = probabilisticForecastingResults["cdfs"]
sampleNum = pdfs.shape[0]
rowNames = []
for i in range(sampleNum):
pdf = pdfs[i]
if i == 0:
pdfsNumpy = np.array([pdf["x"]])
pdfsNumpy = np.r_[pdfsNumpy, [pdf["f"]]]
else:
pdfsNumpy = np.r_[pdfsNumpy, [pdf["x"]]]
pdfsNumpy = np.r_[pdfsNumpy, [pdf["f"]]]
rowNames.append("x")
rowNames.append("f")
pdfsDataFrame = pd.DataFrame(pdfsNumpy, index=rowNames)
PandasInputOutputUtils.savePandasToExcel(pdfsDataFrame, savePath,
"pdfs")
rowNames = []
for i in range(sampleNum):
cdf = cdfs[i]
if i == 0:
cdfsNumpy = np.array([cdf["x"]])
cdfsNumpy = np.r_[cdfsNumpy, [cdf["F"]]]
else:
cdfsNumpy = np.r_[cdfsNumpy, [cdf["x"]]]
cdfsNumpy = np.r_[cdfsNumpy, [cdf["F"]]]
rowNames.append("x")
rowNames.append("F")
cdfsDataFrame = pd.DataFrame(cdfsNumpy, index=rowNames)
PandasInputOutputUtils.savePandasToExcel(cdfsDataFrame, savePath,
"cdfs")
def outputDatasetAfterFeatureOperation(datasetX, datasetY,
featureNamesAfterOperation,
savePath, sheetName):
datasetX = pd.DataFrame(datasetX, columns=featureNamesAfterOperation)
datasetY = pd.DataFrame(datasetY, columns=["L"])
dataset = pd.concat([datasetY, datasetX], axis=1)
PandasInputOutputUtils.savePandasToExcel(dataset, savePath, sheetName)
def outputIntervalForecastingResults(dataset,
probabilisticForecastingResults,
savePath):
if probabilisticForecastingResults is not None:
observations = dataset.validationY.flatten()
observations = dataset.reverseLabel(observations)
observations = observations.flatten()
intervalData = []
rowNames = []
for alpha in [0.8, 0.85, 0.9, 0.95]:
lowerAlpha = (1 - alpha) / 2
upperAlpha = 1 - lowerAlpha
lower = forecast.Forecast.ForecastModelBase.getPredictionsByQuantile(
probabilisticForecastingResults, lowerAlpha)
upper = forecast.Forecast.ForecastModelBase.getPredictionsByQuantile(
probabilisticForecastingResults, upperAlpha)
lower = lower.flatten()
upper = upper.flatten()
intervalData.append(lower)
intervalData.append(upper)
rowNames.append(str(alpha) + "下限")
rowNames.append(str(alpha) + "上限")
intervalData.append(observations)
rowNames.append("真实值")
intervalData = np.array(intervalData).T
intervalDataFrame = pd.DataFrame(intervalData, columns=rowNames)
PandasInputOutputUtils.savePandasToExcel(intervalDataFrame,
savePath, "区间预测结果")
def outputDeterministicForecastingResults(predictions, observations,
savePath):
observationsDataFrame = pd.DataFrame(observations, columns=["真实值"])
predictionsDataFrame = pd.DataFrame(predictions, columns=["预报值"])
allDataFrame = pd.concat([observationsDataFrame, predictionsDataFrame],
axis=1)
PandasInputOutputUtils.savePandasToExcel(allDataFrame, savePath,
"确定性预报结果")
def outputMetricInFeatureSelection(metricValues, featureNames, metricNames,
savePath, sheetName):
if metricValues is not None:
metricNames = [metricName.name for metricName in metricNames]
pandasData = pd.DataFrame(np.array(metricValues).T,
columns=featureNames,
index=metricNames)
PandasInputOutputUtils.savePandasToExcel(pandasData, savePath,
sheetName)
def outputFinalHyperParameters(forecastModel, savePath):
optimizedHyperParameters = forecastModel.getOptimizedHyperParameters()
fixedHyperParameters = forecastModel.getFixedHyperParameters()
optimalParamsFrame = pd.DataFrame(optimizedHyperParameters, index=[0])
PandasInputOutputUtils.savePandasToExcel(optimalParamsFrame, savePath,
"optimizedHyperParameters")
# 采用的固定的超参数
fixedParamsFrame = pd.DataFrame(fixedHyperParameters, index=[0])
PandasInputOutputUtils.savePandasToExcel(fixedParamsFrame, savePath,
"fixedHyperParameters")
def outputFeatureOptimizationProcess(featureOptimizationProcess,
featureNamesAfterOperation, savePath,
sheetName):
if featureOptimizationProcess is not None:
objTrace = featureOptimizationProcess["objTrace"]
varTrace = featureOptimizationProcess["varTrace"]
obj = pd.DataFrame(objTrace, columns=["平均/当前 适应度", "最优/当前 适应度"])
var = pd.DataFrame(varTrace, columns=featureNamesAfterOperation)
dataset = pd.concat([obj, var], axis=1)
PandasInputOutputUtils.savePandasToExcel(dataset, savePath,
sheetName)
def outputHyperParametersOptimizationProcess(
hyperParametersOptimizationProcess, savePath):
if hyperParametersOptimizationProcess is not None:
best = hyperParametersOptimizationProcess["best"]
trials = hyperParametersOptimizationProcess["trials"]
optimizationProcess = trials.results
times = len(optimizationProcess)
# 优化过程中损失值
losses = np.zeros(shape=(times, 1))
bestLosses = np.zeros(shape=(times, 1))
for i in range(times):
optimizationResult = optimizationProcess[i]
losses[i] = optimizationResult['loss']
if i == 0:
bestLosses[i] = losses[i]
else:
if bestLosses[i - 1] <= losses[i]:
bestLosses[i] = bestLosses[i - 1]
else:
bestLosses[i] = losses[i]
lossProcess = np.append(losses, bestLosses, axis=1) # axis=1 横向扩展
lossProcess = pd.DataFrame(lossProcess,
columns=['losses', 'bestLosses'])
PandasInputOutputUtils.savePandasToExcel(lossProcess, savePath,
"lossProcess")
# 超参数组合与损失值对应
colNames = best.keys()
allData = dict()
for colName in colNames:
data = []
for i in range(times):
optimizationResult = optimizationProcess[i]
data.append(optimizationResult['params'][colName])
allData[colName] = data
allDataFrame = pd.DataFrame(allData)
lossesDataFrame = pd.DataFrame(losses[:, 0], columns=['losses'])
allDataFrame = pd.concat([allDataFrame, lossesDataFrame], axis=1)
PandasInputOutputUtils.savePandasToExcel(allDataFrame, savePath,
"hyperParametersProcess")
def testDataset():
partPath = 'resources/test/1518.xlsx'
path = rootPath + partPath
sheetName = "Sheet1"
pandasData = PandasInputOutputUtils.readExcelDataToPandas(path, sheetName)
numpyData = np.array(pandasData)
featureIndexes = np.array(range(2, 10))
operateEnums = np.array([[
FeatureGenerationOperateEnum.Linear, FeatureGenerationOperateEnum.Power
] for _ in range(2, 10)])
transformFuncs = np.array([["y=1*x+0", "y=1*x^2+0"] for _ in range(2, 10)])
historyIndexes = np.array([[i for i in range(3)], [i for i in range(1, 3)],
[i for i in range(3)], [i for i in range(1, 3)],
[i for i in range(1, 3)],
[i for i in range(1, 3)],
[i for i in range(1, 3)],
[i for i in range(1, 3)],
[i for i in range(1, 3)]])
# historyIndexes = np.array([[j for j in range(3)] for _ in range(2, 10)])
labelIndex = 5
print("断点")
def outputForecastingMetrics(metrics, metricNames, savePath, sheetName):
metricDataFrame = pd.DataFrame(np.array(metrics).reshape(1, -1),
columns=metricNames)
PandasInputOutputUtils.savePandasToExcel(metricDataFrame, savePath,
sheetName)