mapeTM,
pd.DataFrame(np.reshape(np.array([altMAPE, mapeGroundTruth]),
newshape=(1, 2)),
columns=['TM', 'GT'])
])
negLLTM = pd.concat([negLLTM, pd.DataFrame([negLL], columns=['TM'])])
lstmExptDir = 'results/nyc_taxi_experiment_continuous_likelihood_noise/'
noiseList = ['0.0', '0.020', '0.040', '0.060', '0.080', '0.10']
negLLLSTM = pd.DataFrame([], columns=['LSTM'])
noiseStrengthLSTM = []
for noise in noiseList:
experiment = lstmExptDir + 'noise' + noise
expResult = ExperimentResult(experiment)
truth = np.concatenate((np.zeros(5333), expResult.truth))
error = np.concatenate((np.zeros(5333), expResult.error))
noiseValue = (truth - data['value']) / data['value']
noiseStrengthLSTM.append(np.std(noiseValue[8000:]))
if not np.isclose(noiseStrengthLSTM[-1], float(noise), rtol=0.1):
print "Warning: Estimated noise strength is different from the given noise"
negLL = np.nanmean(error[startFrom:])
negLLLSTM = pd.concat([negLLLSTM, pd.DataFrame([negLL], columns=['LSTM'])])
lstmExptDir = 'results/nyc_taxi_experiment_continuous_noise/'
noiseList = ['0.0', '0.020', '0.040', '0.060', '0.080', '0.10']
figPath = './result/'
plt.close('all')
# use datetime as x-axis
dataSet = 'nyc_taxi'
filePath = './data/' + dataSet + '.csv'
data = pd.read_csv(filePath,
header=0,
skiprows=[1, 2],
names=['datetime', 'value', 'timeofday', 'dayofweek'])
xaxisDatetime = pd.to_datetime(data['datetime'])
expResultPerturb = ExperimentResult(
'results/nyc_taxi_experiment_continuous_likelihood_perturb/learning_window'
+ str(1001.0) + '/')
negLLLSTM1000Perturb = expResultPerturb.error
truthLSTM1000Perturb = expResultPerturb.truth
expResultPerturb = ExperimentResult(
'results/nyc_taxi_experiment_continuous_likelihood_perturb/learning_window'
+ str(3001.0) + '/')
negLLLSTM3000Perturb = expResultPerturb.error
truthLSTM3000Perturb = expResultPerturb.truth
expResultPerturb = ExperimentResult(
'results/nyc_taxi_experiment_continuous_likelihood_perturb/learning_window'
+ str(6001.0) + '/')
negLLLSTM6000Perturb = expResultPerturb.error
truthLSTM6000Perturb = expResultPerturb.truth
return np.sqrt(np.nanmean(squareDeviation)) / np.nanstd(truth)
def loadExperimentResult(filePath):
expResult = pd.read_csv(filePath,
header=0,
skiprows=[1, 2],
names=['step', 'value', 'prediction5'])
groundTruth = np.roll(expResult['value'], -5)
prediction5step = np.array(expResult['prediction5'])
return (groundTruth, prediction5step)
if __name__ == "__main__":
xaxisDate = getDatetimeAxis()
expResult = ExperimentResult(
'results/nyc_taxi_experiment_continuous/learning_window6001.0/')
# ### Figure 1: Continuous vs Batch LSTM
# fig = plt.figure()
# # NRMSE_StaticLSTM = plotLSTMresult('results/nyc_taxi_experiment_one_shot/',
# # window, xaxis=xaxis_datetime, label='static lstm')
# (nrmseLSTM6000, expResultLSTM6000) = \
# plotLSTMresult('results/nyc_taxi_experiment_continuous/learning_window6001.0/',
# window, xaxis=xaxisDate, label='continuous LSTM-6000')
# plt.legend()
# plt.savefig(figPath + 'continuousVsbatch.pdf')
#
### Figure 2: Continuous LSTM with different window size
fig = plt.figure()
figPath = './result/'
plt.close('all')
# use datetime as x-axis
dataSet = 'nyc_taxi'
filePath = './data/' + dataSet + '.csv'
data = pd.read_csv(filePath,
header=0,
skiprows=[1, 2],
names=['datetime', 'value', 'timeofday', 'dayofweek'])
xaxis_datetime = pd.to_datetime(data['datetime'])
expResult_perturb = ExperimentResult(
'results/nyc_taxi_experiment_continuous_likelihood_perturb/learning_window'
+ str(1001.0) + '/')
negLL_LSTM1000_perturb = expResult_perturb.error
truth_LSTM1000_perturb = expResult_perturb.truth
expResult_perturb = ExperimentResult(
'results/nyc_taxi_experiment_continuous_likelihood_perturb/learning_window'
+ str(3001.0) + '/')
negLL_LSTM3000_perturb = expResult_perturb.error
truth_LSTM3000_perturb = expResult_perturb.truth
expResult_perturb = ExperimentResult(
'results/nyc_taxi_experiment_continuous_likelihood_perturb/learning_window'
+ str(6001.0) + '/')
negLL_LSTM6000_perturb = expResult_perturb.error
truth_LSTM6000_perturb = expResult_perturb.truth