from Dataset import Dataset from Discova import Discova data = Dataset() data.load_house_dataframe(1) data.add_windows(1, '00 mains') data.add_windows(1, '05 refrigerator') X_train, X_val, Y_train, Y_val = data.format_for_keras(1, '05 refrigerator') ''' discova = Discova() discova.load_data(X_train, X_val, Y_train, Y_val) discova.construct_model() discova.compile_model() discova.load_tensorboard() discova.train_model() '''
for train_house in [5]:
discova = Autoencoder() # Turn off variance for prediction
discova.construct_model()
discova.compile_model()
path = 'weights/ae_0' + str(train_house) + '_washerdryer.hdf5'
discova.model.load_weights(path)
print(('Loading weights for House ' + str(train_house)).ljust(40,'.'))
for test_house in range(1,7):
load_str = 'Loading test data for House ' + str(test_house)
print(load_str.ljust(40,'.'))
applianceLabel = applianceLabels[test_house - 1]
data = Dataset()
data.load_house_dataframe(test_house)
data.add_windows(test_house, '00 mains')
data.add_windows(test_house, applianceLabel)
data.add_statistics(test_house)
print('Predicting'.ljust(40,'.'))
predRaw = discova.model.predict(data.windows[test_house]['00 mains'])
print('Aligning by timestep'.ljust(40,'.'))
predTimestep = data.recover_reverse_diagonals(predRaw)
print('Taking median'.ljust(40,'.'))
predMedian = np.median(predTimestep, axis = 1)
print('Rescaling'.ljust(40,'.'))
mean = data.means[test_house][applianceLabel]
std = data.stddevs[test_house][applianceLabel]
predMedianScaled = predMedian * std + mean
print('Calculating relative MAE'.ljust(40,'.'))
true = data.dataframes[test_house][applianceLabel][511:-511].as_matrix()
mae = np.abs(true - predMedianScaled)
def evaluateInHouse(self, networkType):
for appliance in self.applianceTypes:
print('Adding results for {}'.format(appliance).ljust(40, '.'))
for house in self.houses:
try:
network = self.getNetworkWithWeights(
networkType, appliance, house)
load_str = 'Loading test data for House ' + str(house)
print(load_str.ljust(40, '.'))
data = Dataset()
data.load_house_dataframe(house)
data.add_windows(house, '00 mains')
mainsWindows = self.getMainsWindows(data,
house,
outOfHouse=False)
print('Predicting'.ljust(40, '.'))
predRaw = network.model.predict(mainsWindows)
print('Aligning by timestep'.ljust(40, '.'))
predTimestep = data.recover_reverse_diagonals(predRaw)
print('Taking median'.ljust(40, '.'))
predMedian = np.median(predTimestep, axis=1)
print('Rescaling'.ljust(40, '.'))
data.add_statistics(house)
applianceLabel = self.dfLabels.loc[house, appliance]
mean = data.means[house][applianceLabel]
std = data.stddevs[house][applianceLabel]
predMedianScaled = predMedian * std + mean
print('Calculating relative MAE'.ljust(40, '.'))
data.add_windows(house, applianceLabel)
m = data.dataframes[house].shape[0]
testStartIdx = int(-0.2 * m)
true = data.dataframes[house][applianceLabel]
true = true[testStartIdx:-511].as_matrix()
mae = np.abs(true - predMedianScaled)
relative_mae = mae.sum() / true.sum()
print('Appending results'.ljust(40, '.'))
row_dict = {
'network': networkType,
'appliance': appliance,
'train_house': house,
'predict_house': house,
'relative_mae': relative_mae
}
except Exception:
traceback.print_exc()
row_dict = {
'network': networkType,
'appliance': appliance,
'train_house': house,
'predict_house': house,
'relative_mae': None
}
print('Error on House {} {}'.format(house,
appliance).ljust(
40, '.'))
self.results = self.results.append(row_dict, ignore_index=True)
def evaluate_latent_space(self, networkType):
for appliance in self.applianceTypes:
print('Adding results for {}'.format(appliance).ljust(40, '.'))
for house in self.houses:
try:
network = self.getNetworkWithWeights(
networkType, appliance, house)
latent_model = Model(
inputs=network.model.input,
outputs=network.model.layers[13].output)
load_str = 'Loading test data for House ' + str(house)
print(load_str.ljust(40, '.'))
data = Dataset()
data.load_house_dataframe(house)
data.add_windows(house, '00 mains')
mainsWindows = self.getMainsWindows(data,
house,
outOfHouse=False)
print('Getting latent vectors'.ljust(40, '.'))
latent_vecs = latent_model.predict(mainsWindows)
print('Calculating neighbor distance'.ljust(40, '.'))
neighbor_diff = latent_vecs[1:] - latent_vecs[:-1]
mean_latent_dist = np.linalg.norm(neighbor_diff,
axis=1).mean()
print('Calculating neighbor cosine similarity'.ljust(
40, '.'))
mean_cosine_sim = paired_cosine_distances(
latent_vecs[1:], latent_vecs[:-1]).mean()
print('Appending results'.ljust(40, '.'))
row_dict = {
'network': networkType,
'appliance': appliance,
'train_house': house,
'predict_house': house,
'euclidean_dist': mean_latent_dist,
'cosine_similarity': mean_cosine_sim
}
except Exception:
traceback.print_exc()
row_dict = {
'network': networkType,
'appliance': appliance,
'train_house': house,
'predict_house': house,
'euclidean_dist': None,
'cosine_similarity': None
}
print('Error on House {} {}'.format(house,
appliance).ljust(
40, '.'))
self.results = self.results.append(row_dict, ignore_index=True)
#latent = LatentSpace()
#latent.evaluate_latent_space('vae')
