nlpd_samples, nlpd_meanvar = m.nlpd_samples(test.X,
test.Y,
batch_size=BATCH_SIZE)
try:
np.savetxt("nlpd_meanvar.csv", nlpd_meanvar,
delimiter=",") # N x 2P as for predictions
except:
print('nlpd_meanvar export fail')
#try:
# np.savetxt("nlpd_samples.csv", nlpd_samples, delimiter=",") # NP x S (NxS concat for P tasks)
#except:
# print('nlpd_samples export fail')
print("Final " + error_rate.get_name() + "=" +
"%.4f" % error_rate.eval(test.Y, ypred[0]))
print("Final " + "generalised_nlpd" + "=" +
"%.4f" % m.nlpd_general(test.X, test.Y, batch_size=BATCH_SIZE))
error_rate_end = [losses.MeanAbsError(data.Dout)
] # any extra accuracy measures at end of routine
print("Final ", [e.get_name() for e in error_rate_end])
print([e.eval(test.Y, ypred[0]) for e in error_rate_end])
predvar = [np.mean(np.mean(ypred[1]))]
print("Final predvar ", predvar)
with open("final_losses.csv", 'w', newline='') as f:
csv.writer(f).writerows(
[[e.get_name() for e in error_rate_end] + ['pred_var'],
[e.eval(test.Y, ypred[0]) for e in error_rate_end] + predvar])
print("finish time = " + time.strftime('%X %x %Z'))
# export final predicted values and loss metrics
ypred = m.predict(test.X, batch_size = BATCH_SIZE) #same batchsize used for convenience
np.savetxt("predictions.csv", np.concatenate(ypred, axis=1), delimiter=",")
if save_nlpds == True:
nlpd_samples, nlpd_meanvar = m.nlpd_samples(test.X, test.Y, batch_size = BATCH_SIZE)
try:
np.savetxt("nlpd_meanvar.csv", nlpd_meanvar, delimiter=",") # N x 2P as for predictions
except:
print('nlpd_meanvar export fail')
try:
np.savetxt("nlpd_samples.csv", nlpd_samples, delimiter=",") # NP x S (NxS concat for P tasks)
except:
print('nlpd_samples export fail')
print("Final " + error_rate.get_name() + "=" + "%.4f" % error_rate.eval(test.Y, ypred[0]))
print("Final " + "generalised_nlpd" + "=" + "%.4f" % m.nlpd_general(test.X, test.Y, batch_size = BATCH_SIZE))
error_rate_end = [losses.MeanAbsError(data.Dout)] # any extra accuracy measures at end of routine
print("Final ", [e.get_name() for e in error_rate_end])
print([e.eval(test.Y, ypred[0]) for e in error_rate_end])
predvar = [np.mean(np.mean(ypred[1]))]
print("Final predvar ", predvar)
with open("final_losses.csv", 'w', newline='') as f:
csv.writer(f).writerows([[e.get_name() for e in error_rate_end] + ['pred_var'],
[e.eval(test.Y, ypred[0]) for e in error_rate_end] + predvar])
print("finish time = " + time.strftime('%X %x %Z'))
