def performEvaluations(filename,
train_window=3000,
overload_dur=5,
overload_percentile=70,
steps=30):
cur_results = []
forecasts = np.nan_to_num(
np.genfromtxt("d:/data/" + TYPE + "_" + METHOD + "/" + filename,
delimiter=',',
usecols=range(0, 30))).ravel() # ,usecols=range(0,30)
truevals = np.genfromtxt("d:/data/" + TYPE + "/" + filename,
delimiter=',',
skip_header=1)[:train_window + len(forecasts), 1]
# Normalize
# truevals = np.divide(truevals, np.max(truevals))
threshold = np.percentile(truevals, overload_percentile)
cur_results.append(eval.calc_RMSE(truevals[train_window:], forecasts))
for val in eval.calc_upper_lower_acc(truevals[train_window:], forecasts):
cur_results.append(val)
for val in eval.calc_persample_accuracy(truevals[train_window:], forecasts,
threshold):
cur_results.append(val)
for val in eval.calc_overload_states_acc(truevals[train_window:],
forecasts, threshold):
cur_results.append(val)
return cur_results
def trainFunc(params):
iter, trainds, validds, input_size, hidden, func, eta, lmda, epochs = params
print('Iter:', iter, 'Epochs:', epochs, 'Hidden_size:', hidden, 'Eta:',
eta, 'Lamda:', lmda, 'Activation:', func)
# Build network
n = RecurrentNetwork()
n.addInputModule(LinearLayer(input_size, name='in'))
n.addModule(func(hidden, name='hidden'))
n.addModule(LinearLayer(hidden, name='context'))
n.addOutputModule(LinearLayer(1, name='out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name='in_to_hidden'))
n.addConnection(FullConnection(n['hidden'], n['out'],
name='hidden_to_out'))
n.addRecurrentConnection(FullConnection(n['hidden'], n['context']))
rnet = n
rnet.sortModules()
trainer = BackpropTrainer(n,
trainds,
learningrate=eta,
weightdecay=lmda,
momentum=0.1,
shuffle=False)
trainer.trainEpochs(epochs)
pred = np.nan_to_num(n.activateOnDataset(validds))
validerr = eval.calc_RMSE(validds['target'], pred)
varscore = explained_variance_score(validds['target'], pred)
return validerr, varscore, n
def performEvaluations(params, train_window = 3000, overload_dur = 5, overload_percentile = 70, predic_window=30):
filename, METHOD, TYPE, OUTPUT, INPUT = params[:5]
filename = filename.split('/')[-1]
print OUTPUT+TYPE+"_"+METHOD+"/" + filename, "started..."
cur_results = []
forecasts = np.nan_to_num(np.genfromtxt(OUTPUT+TYPE+"_"+METHOD+"/" + filename, delimiter=',',usecols=range(0,predic_window))).ravel() # ,usecols=range(0,30)
if TYPE == 'pageviews' or TYPE == 'network':
filename = filename.replace(".csv","")
truevals = np.genfromtxt(INPUT+TYPE+"/"+filename)[:train_window+len(forecasts)]
truevals = truevals/np.max(truevals)
else:
truevals = np.genfromtxt(INPUT+TYPE+"/"+filename, delimiter=',',skip_header=1)[:train_window+len(forecasts),1]
# Normalize
# truevals = np.divide(truevals, np.max(truevals))
threshold = np.percentile(truevals, overload_percentile)
cur_results.append(eval.calc_RMSE(truevals[train_window:], forecasts))
for val in eval.calc_upper_lower_acc(truevals[train_window:], forecasts):
cur_results.append(val)
for val in eval.calc_persample_accuracy(truevals[train_window:], forecasts, threshold):
cur_results.append(val)
for val in eval.calc_overload_states_acc(truevals[train_window:], forecasts, threshold):
cur_results.append(val)
return cur_results
def performEvaluations(filename,
train_window=3000,
overload_dur=5,
overload_percentile=70,
steps=30):
cur_results = []
forecasts = np.genfromtxt("d:/data/cpu_norm_forecasts/" + filename,
delimiter=',',
usecols=range(0, steps)).ravel()
truevals = np.genfromtxt("d:/data/cpuRate/" + filename,
delimiter=',',
skip_header=1)[train_window:train_window +
len(forecasts), 1]
threshold = np.percentile(truevals, overload_percentile)
cur_results.append(eval.calc_RMSE(truevals, forecasts))
for val in eval.calc_upper_lower_acc(truevals, forecasts):
cur_results.append(val)
for val in eval.calc_persample_accuracy(truevals, forecasts, threshold):
cur_results.append(val)
for val in eval.calc_overload_states_acc(truevals, forecasts, threshold):
cur_results.append(val)
return cur_results
def trainFunc(params):
iter, trainds, validds, input_size, hidden, func, eta, lmda, epochs = params
print("Iter:", iter, "Epochs:", epochs, "Hidden_size:", hidden, "Eta:", eta, "Lamda:", lmda, "Activation:", func)
net = buildNetwork(input_size, hidden, 1, bias=True)
trainer = BackpropTrainer(net, trainds, learningrate=eta, weightdecay=lmda, momentum=0.1, shuffle=False)
trainer.trainEpochs(epochs)
pred = np.nan_to_num(net.activateOnDataset(validds))
validerr = eval.calc_RMSE(validds["target"], pred)
varscore = explained_variance_score(validds["target"], pred)
return validerr, varscore, net
def trainFunc(params):
iter, trainds, validds, input_size, hidden, func, eta, lmda, epochs = params
print('Iter:', iter, 'Epochs:', epochs, 'Hidden_size:', hidden, 'Eta:', eta, 'Lamda:', lmda, 'Activation:', func)
net = buildNet(input_size, hidden)
trainer = BackpropTrainer(net, trainds, learningrate=eta, weightdecay=lmda, momentum=0.1, shuffle=False)
trainer.trainEpochs(epochs)
pred = np.nan_to_num(net.activateOnDataset(validds))
validerr = eval.calc_RMSE(validds['target'], pred)
varscore = explained_variance_score(validds['target'], pred)
return validerr, varscore, net
def performEvaluations(filename, train_window = 3000, overload_dur = 5, overload_percentile = 70, steps=30):
cur_results = []
filename
forecasts = np.nan_to_num(np.genfromtxt("d:/Wikipage data/"+TYPE+"_"+METHOD+"/" + filename, delimiter=',',usecols=range(0,30))).ravel() #
# truevals = np.genfromtxt("d:/Wikipage data/"+TYPE+"/"+filename, delimiter=',',skip_header=1)[:train_window+len(forecasts),1]
truevals = np.genfromtxt("d:/Wikipage data/"+TYPE+"/"+filename)[:train_window+len(forecasts)]
truevals = truevals/np.max(truevals)
cur_results.append(eval.calc_RMSE(truevals[train_window:], forecasts))
for val in eval.calc_upper_lower_acc(truevals[train_window:], forecasts):
cur_results.append(val)
return cur_results
def performEvaluations(filename, train_window = 3000, overload_dur = 5, overload_percentile = 70, steps=30):
cur_results = []
forecasts = np.genfromtxt("d:/data/diskio_ar_forecasts/"+ filename,delimiter=',',usecols=range(0,steps)).ravel()
truevals = np.genfromtxt("d:/data/diskio/"+filename, delimiter=',',skip_header=1)[train_window:train_window+len(forecasts),1]
threshold = np.percentile(truevals, overload_percentile)
cur_results.append(eval.calc_RMSE(truevals, forecasts))
for val in eval.calc_upper_lower_acc(truevals, forecasts):
cur_results.append(val)
for val in eval.calc_persample_accuracy(truevals, forecasts, threshold):
cur_results.append(val)
for val in eval.calc_overload_states_acc(truevals, forecasts, threshold):
cur_results.append(val)
return cur_results
def trainFunc(params):
iter, trainds, validds, input_size, hidden, func, eta, lmda, epochs = params
print('Iter:', iter, 'Epochs:', epochs, 'Hidden_size:', hidden, 'Eta:',
eta, 'Lamda:', lmda, 'Activation:', func)
net = buildNetwork(input_size, hidden, 1, bias=True)
trainer = BackpropTrainer(net,
trainds,
learningrate=eta,
weightdecay=lmda,
momentum=0.1,
shuffle=False)
trainer.trainEpochs(epochs)
pred = np.nan_to_num(net.activateOnDataset(validds))
validerr = eval.calc_RMSE(validds['target'], pred)
varscore = explained_variance_score(validds['target'], pred)
return validerr, varscore, net
def performEvaluations(filename, train_window = 3000, overload_dur = 5, overload_percentile = 70, steps=30):
cur_results = []
forecasts = np.nan_to_num(np.genfromtxt("d:/data/"+TYPE+"_"+METHOD+"/" + filename, delimiter=',',usecols=range(0,30))).ravel() # ,usecols=range(0,30)
truevals = np.genfromtxt("d:/data/"+TYPE+"/"+filename, delimiter=',',skip_header=1)[:train_window+len(forecasts),1]
# Normalize
# truevals = np.divide(truevals, np.max(truevals))
threshold = np.percentile(truevals, overload_percentile)
cur_results.append(eval.calc_RMSE(truevals[train_window:], forecasts))
for val in eval.calc_upper_lower_acc(truevals[train_window:], forecasts):
cur_results.append(val)
for val in eval.calc_persample_accuracy(truevals[train_window:], forecasts, threshold):
cur_results.append(val)
for val in eval.calc_overload_states_acc(truevals[train_window:], forecasts, threshold):
cur_results.append(val)
return cur_results
def trainFunc(params):
iter, trainds, validds, input_size, hidden, func, eta, lmda, epochs = params
print('Iter:', iter, 'Epochs:', epochs, 'Hidden_size:', hidden, 'Eta:', eta, 'Lamda:', lmda, 'Activation:', func)
# Build network
n = RecurrentNetwork()
n.addInputModule(LinearLayer(input_size, name = 'in'))
n.addModule(func(hidden, name = 'hidden'))
n.addModule(LinearLayer(hidden, name = 'context'))
n.addOutputModule(LinearLayer(1, name = 'out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name = 'in_to_hidden'))
n.addConnection(FullConnection(n['hidden'], n['out'], name = 'hidden_to_out'))
n.addRecurrentConnection(FullConnection(n['hidden'], n['context']))
rnet = n
rnet.sortModules()
trainer = BackpropTrainer(n, trainds, learningrate=eta, weightdecay=lmda, momentum=0.1, shuffle=False)
trainer.trainEpochs(epochs)
pred = np.nan_to_num(n.activateOnDataset(validds))
validerr = eval.calc_RMSE(validds['target'], pred)
varscore = explained_variance_score(validds['target'], pred)
return validerr, varscore, n
def performEvaluations(filename,
train_window=3000,
overload_dur=5,
overload_percentile=70,
steps=30):
cur_results = []
filename
forecasts = np.nan_to_num(
np.genfromtxt("d:/Wikipage data/" + TYPE + "_" + METHOD + "/" +
filename,
delimiter=',',
usecols=range(0, 30))).ravel() #
# truevals = np.genfromtxt("d:/Wikipage data/"+TYPE+"/"+filename, delimiter=',',skip_header=1)[:train_window+len(forecasts),1]
truevals = np.genfromtxt("d:/Wikipage data/" + TYPE + "/" +
filename)[:train_window + len(forecasts)]
truevals = truevals / np.max(truevals)
cur_results.append(eval.calc_RMSE(truevals[train_window:], forecasts))
for val in eval.calc_upper_lower_acc(truevals[train_window:], forecasts):
cur_results.append(val)
return cur_results
def ensembleModel(params, types=['ma','ar','fnn','agile'], step=30, input_window=3000):
input_size = len(types)
filename, METHOD, TYPE, OUTPUT = params[0:4]
filename = filename.split('/')[-1]
filename, METHOD, TYPE, OUTPUT = params[0:4]
filename = filename.split('/')[-1]
combine_model = np.genfromtxt(OUTPUT+TYPE+"_"+types[0]+"/"+filename, delimiter=',', usecols=range(0,30)).ravel()
truevals = np.nan_to_num(np.genfromtxt(OUTPUT+TYPE+"/"+filename, delimiter=',',skip_header=1, usecols=(1))[:input_window+len(combine_model)])
for t in types[1:]:
forecasts = np.genfromtxt(OUTPUT+TYPE+"_"+t+"/"+filename, delimiter=',', usecols=range(0,30)).ravel()
combine_model = np.vstack((combine_model, forecasts))
average_fc = np.average(combine_model, axis=0)
if METHOD == 'avg4':
fileutils.writeCSV(OUTPUT+TYPE+"_"+METHOD+"/"+filename, np.atleast_2d(average_fc).reshape([178,30]))
print filename, "complete"
return
if METHOD == 'combo4' or METHOD =='wa':
training = SupervisedDataSet(input_size, 1)
for i in range(step):
training.appendLinked([combine_model[t][i] for t in range(input_size)], truevals[i+input_window])
besterr = eval.calc_RMSE(truevals[input_window:input_window+step], average_fc[:step])
bestNet = None
for i in range(50):
if METHOD == 'wa':
net = buildNetwork(input_size, 1, hiddenclass=LinearLayer, bias=False)
else:
net = buildNetwork(input_size, 2, 1, hiddenclass=LinearLayer, bias=False)
trainer = BackpropTrainer(net, training, learningrate=0.001, shuffle=False)
trainer.trainEpochs(100)
err = eval.calc_RMSE(truevals[input_window:input_window+step], net.activateOnDataset(training))
if err < besterr:
bestNet = net
break
combo_fc = average_fc[0:step].tolist()
# combo_fc = []
if bestNet == None:
combo_fc = average_fc
else:
for i in range(step, len(combine_model[0]), step):
training = SupervisedDataSet(input_size, 1)
for j in range(i,i+step):
combo_fc.append(bestNet.activate([combine_model[t][j] for t in range(input_size)])[0])
training.appendLinked([combine_model[t][j] for t in range(input_size)], truevals[j+input_window])
trainer = BackpropTrainer(bestNet, training, learningrate=0.01, shuffle=False)
trainer.trainEpochs(2)
result = np.atleast_2d(combo_fc).reshape([178,30])
minimum = np.percentile(truevals,5)
result[0,result[0,:] < minimum] = minimum
fileutils.writeCSV(OUTPUT+TYPE+"_"+METHOD+"/"+filename, result)
print filename, "complete"
