rmseLevels = {}
maeLevels = {}
rLevels = {}
fac2Levels = {}
nmseLevels = {}
fbLevels = {}
rsLevels = {}
mgLevels = {}
vgLevels = {}
for method in methods:
print("Method: " + method)
rmse = rmseEval(observations[method], predictions[method])[1]
print("\trmse: " + str(rmse))
mae = maeEval(observations[method], predictions[method])[1]
print("\tmae: " + str(mae))
r = correlationEval(observations[method], predictions[method])[1]
print("\tr: " + str(r))
print("\tr2: " +
str(rsquaredEval(observations[method], predictions[method])[1]))
print("\tr2: " + str(r2_score(observations[method], predictions[method])))
fac2 = fac2Eval(observations[method], predictions[method])
print("\tfac2: " + str(fac2))
print("\tmg: " + str(mgEval(observations[method], predictions[method])))
nmse = nmse_from_paper(observations[method], predictions[method])
print("\tnmse: " + str(nmse))
fb = fbEval(observations[method], predictions[method])[1]
print("\tfb: " + str(fb))
rs = evalRandomScatter(observations[method], predictions[method])
print("\trandom_scatter: " + str(rs))
modelledData.append(float('nan'))
d = datetime(timestamp.year, timestamp.month, timestamp.day, timestamp.hour)
datesList.append(d)
if timestamp.key not in aqData or str(loc) not in aqData[timestamp.key]:
aqDataStation.append(NAN)
else:
aqDataStation.append(aqData[timestamp.key][str(loc)])
observed.append(aqData[timestamp.key][str(loc)])
if timestamp.key in finalData and loc in finalData[timestamp.key]:
predicted.append(finalData[timestamp.key][loc])
else:
predicted.append(float('nan'))
if len(predicted) > 0:
rmse = rmseEval(predicted, observed)
mae = maeEval(predicted, observed)
#r = correlationEval(predicted, observed)
# print("r: " + str(r))
# print("RMSE: " + str(rmse))
# print("MAE: " + str(mae))
fig = plt.figure(None, figsize=(15, 9))
ax = fig.add_subplot(111)
hfmt = dates.DateFormatter('%d/%m/%Y %H:%M')
ax.xaxis.set_major_formatter(hfmt)
if len(predicted) > 0:
ax.plot_date(datesList, modelledData, '-', color="b", label="Modelled (RMSE:" + str(rmse[1])[0:5] + ",MAE:" + str(mae[1])[0:5] + ")")
else:
ax.plot_date(datesList, modelledData, '-', color="b", label="Modelled")
ax.plot_date(datesList, aqDataStation, '-', color="r", label="Observed")
ax.legend()
for v in values:
sName = stationNames[str(v)]
print("location: " + str(v) + " -> " + sName)
trainData, testData = splitDataForXValidation(v, "location", data)
trainColumns = []
for c in trainData:
if c != "target":
trainColumns.append(c)
model = trainRandomForest(trainData, trainColumns, "target", {
'estimators': 59,
'leaf': 9
})
predictionData = applyRandomForest(testData, model, 0)
eval1 = maeEval(testData["target"], predictionData)
eval2 = maeDistribution(testData["target"], predictionData, 50)
eval3 = maeDistribution2(testData["target"], predictionData, 50)
eval4 = reDistribution(testData["target"], predictionData)
eval5 = reDistribution2(testData["target"], predictionData)
doGraph(eval2[1], "No2 hourly AE (MAE: " + str(eval1[1]) + ") @ " + sName,
"AE (ug/m3)", OUTPUT_DIRECTORY + "ae_" + sName + ".png")
doGraph2(50, eval3[1],
"No2 hourly error (MAE: " + str(eval1[1]) + ") @ " + sName,
"error (ug/m3)", OUTPUT_DIRECTORY + "e_" + sName + ".png")
doGraph3(eval4[1], "No2 hourly RAE @ " + sName, "absolute relative error",
OUTPUT_DIRECTORY + "rae_" + sName + ".png")
doGraph4(eval5[1], "No2 hourly RE @ " + sName, "relative error",
OUTPUT_DIRECTORY + "re_" + sName + ".png")
for i in range(0, len(eval2[1])):
predictions = []
for timestamp in ospm:
if timestamp not in obs:
continue
pred = ospm[timestamp]
o = obs[timestamp]
observations.append(o)
predictions.append(pred)
c = c + 1
obs_avg = obs_avg + o
pred_avg = pred_avg + pred
pred_avg = pred_avg / float(c)
obs_avg = obs_avg / float(c)
print("location: " + str(station))
print("\tcounter: " + str(c))
print("\tpred_avg: " + str(pred_avg))
print("\tobs_avg: " + str(obs_avg))
observationPerStation.append(obs_avg)
predictionPerStation.append(pred_avg)
rmse = rmseEval(observations, predictions)[1]
print("\tRMSE: " + str(rmse))
rmse = rmseEval(observationPerStation, predictionPerStation)[1]
print("RMSE: " + str(rmse))
mae = maeEval(observationPerStation, predictionPerStation)[1]
print("MAE: " + str(mae))