def __init__(self, a, d, p):
self.aParamList = a
self.dParamList = d
self.pParamList = p
#Output plot
self.plot = outPlot.OutputTimeSeriesPlot("Gammafunctions.html",
"Gamma function",
"with different parameters",
"f(x)", "")
# Read the data
df = pd.read_csv('BMW_comments_interaction.csv',
index_col=0,
parse_dates=True,
names=['value'])
# Convert the dataframe into series
series = pd.Series(data=np.array(df.as_matrix()).flatten(), index=df.index)
resampled_series = series.resample(rule='H', how=_sum)
xAxis = []
yAxis = []
for index, value in resampled_series.iteritems():
year = int(index.year)
month = int(index.month) - 1
day = int(index.day)
hour = int(index.hour)
nextDayStr = "Date.UTC(" + str(year) + "," + str(
int(month)) + "," + str(day) + "," + str(hour) + ")"
xAxis.append(nextDayStr)
yAxis.append(str(value))
# Plotting of the actual and prediction output
outputFolderName = "Outputs/Outputs-Comments-Hourly" + str(datetime.now())
os.mkdir(outputFolderName)
outplot = plotting.OutputTimeSeriesPlot(outputFolderName + "/OwnPosts.html",
"Facebook Comments-BMW", "",
"Number of Comments")
outplot.setSeries('Own Posts', np.array(xAxis), np.array(yAxis))
outplot.createOutput()
month = nextDate.strftime("%m")
day = nextDate.strftime("%d")
nextDayStr = "Date.UTC(" + year+","+ str((int(month)-1)) + "," + day +")"
xAxis.append(nextDayStr)
#Form the feature vectors
feature = [1.0]
for interval in featureIntervalList:
feature.append(series[nextDate + interval])
feature = np.array(feature).reshape((1,len(featureIntervalList)+1))
predictedValue = res2.predict(feature)[0,0]
predicted.append(predictedValue)
#Add it to the series
series[nextDate] = predictedValue
predicted2 = minMax.inverse_transform(np.array(predicted))
# Plotting of the actual and prediction output
outputFolderName = "Outputs/Outputs" + str(datetime.now()) + "_depth_" + str(depth) + "_horizon_" + str(horizon)
os.mkdir(outputFolderName)
outplot = outTimePlot.OutputTimeSeriesPlot(outputFolderName + "/Prediction.html", "Facebook Own posts-BMW", "", "Number of posts")
outplot.setSeries('Actual Output', np.array(xAxis), actualData)
outplot.setSeries('Predicted Output 1', np.array(xAxis), predicted1)
outplot.setSeries('Predicted Output 2', np.array(xAxis), predicted2)
outplot.createOutput()
predictedDict[str(depth)] = deScaled
#Calculate the error
errorFunction = rmse.MeanSquareError()
error = errorFunction.compute(actualData.reshape(1, horizon),
deScaled.reshape(1, horizon))
regressionError.append(error)
xAxisError.append("Depth_" + str(depth))
# Plotting of the actual and prediction output
outputFolderName = "Outputs/Output" + str(
datetime.now()) + "_depth_comparison_" + "_horizon_" + str(horizon)
os.mkdir(outputFolderName)
outplot = outTimePlot.OutputTimeSeriesPlot(
outputFolderName + "/Prediction.html", "Comparison of prediction outputs",
"with varying depths", "Likes Count")
outplot.setSeries('Actual Output', np.array(xAxis), actualData)
for depth in depthList:
seriesName = 'Predicted_Output_depth' + str(depth)
predicted = predictedDict[str(depth)]
outplot.setSeries(seriesName, np.array(xAxis), predicted)
outplot.createOutput()
#Plotting of regression
errPlot = errorPlot.ErrorPlot(outputFolderName + "/Regression_Error.html",
"Comparison of regression error",
"with varying depths", "ESN Configuration",
"Total Error")
#X-axis
errPlot.setXAxis(np.array(xAxisError))
# Read the data
df = pd.read_csv('facebookPosts_ferrari_time_photo_interaction.csv',
index_col=0,
parse_dates=True,
names=['value'])
# Convert the dataframe into series
series = pd.Series(data=np.array(df.as_matrix()).flatten(), index=df.index)
resampled_series = series.resample(rule='H', how=_sum)
xAxis = []
yAxis = []
for index, value in resampled_series.iteritems():
year = int(index.year)
month = int(index.month) - 1
day = int(index.day)
hour = int(index.hour)
nextDayStr = "Date.UTC(" + str(year) + "," + str(
int(month)) + "," + str(day) + "," + str(hour) + ")"
xAxis.append(nextDayStr)
yAxis.append(str(value))
# Plotting of the actual and prediction output
outputFolderName = "Outputs/Outputs-OwnPosts-Hourly" + str(datetime.now())
os.mkdir(outputFolderName)
outplot = plotting.OutputTimeSeriesPlot(
outputFolderName + "/OwnPosts.html",
"Facebook Own Posts Interaction-Ferrari", "", "Interaction")
outplot.setSeries('Own Posts', np.array(xAxis), np.array(yAxis))
outplot.createOutput()
numberOfOUtliers = 0
for index, value in actualSeries.iteritems():
year = index.strftime("%Y")
month = index.strftime("%m")
day = index.strftime("%d")
hour = index.strftime("%H")
nextDayStr = "Date.UTC(" + str(year) + "," + str(int(month) -
1) + "," + str(day) + ")"
xAxis.append(nextDayStr)
if (outlierSeries[index] == 1):
color.append("red")
radius.append(10)
numberOfOUtliers += 1
else:
color.append("blue")
radius.append(4)
# Step 4 - Plot the actual data
fileName = "/Outlier_Detection_Using_STD.html"
actualDataPlot = plot.OutputTimeSeriesPlot(
outputFolderName + fileName, "Facebook Fans Change",
"Outlier Detection using standard deviation approach", "Fans Change")
actualDataPlot.setSeriesWithColorAndRadius("Fans Change", np.array(xAxis),
actualSeries.values.flatten(),
np.array(color), np.array(radius))
actualDataPlot.createOutput()
print("Number of outliers: " + str(numberOfOUtliers))
#Compose the query
query = [1.0]
for d in range(1, depth + 1):
query.append(availableData[nextDayIndex - d, 3])
nextDayPredicted = res.predict(np.array(query).reshape((1, depth + 1)))
predictedOutput.append(nextDayPredicted[0, 0])
#Update the available data
availableData[nextDayIndex, 3] = nextDayPredicted[0, 0]
availableData[nextDayIndex, 0] = int(year)
availableData[nextDayIndex, 1] = int(month)
availableData[nextDayIndex, 2] = int(day)
lastDay = nextDay
lastDayIndex = nextDayIndex
predicted = minMax.inverse_transform(np.array(predictedOutput))
# Plotting of the actual and prediction output
outputFolderName = "Outputs" + str(
datetime.now()) + "_depth_" + str(depth) + "_horizon_" + str(horizon)
os.mkdir(outputFolderName)
outplot = outTimePlot.OutputTimeSeriesPlot(
outputFolderName + "/Prediction.html", "Likes count for facebook page-BMW",
"", "Likes Count")
outplot.setSeries('Actual Output', np.array(xAxis), actualData)
outplot.setSeries('Predicted Output', np.array(xAxis), predicted)
outplot.createOutput()