def plot_queries(stats: TrialStats):
stats.plot('queries', 'a=5')
stats.plot('queries', 'a=7')
stats.plot('queries', 'a=9')
plt.title('Queries')
plt.legend()
plt.show()
def plot_accuracies(stats: TrialStats):
stats.plot('accuracies', 'a=5')
stats.plot('accuracies', 'a=7')
stats.plot('accuracies', 'a=9')
plt.title('Accuracy')
plt.legend()
plt.show()
def accuracy_vs_time_vs_appliances_func(seed: int,
house_trials: int,
duration: int,
a: int,
house_pool: Sequence[int],
matcher_spec: MatcherSpec,
keys={}):
trial_stats = TrialStats(keys=keys)
reconstruct_settings = data_combiner.ReconstructionSettings(
duration, -1, a)
for house_num in house_pool:
for i in range(house_trials):
print('\ta={},h={},i={}'.format(a, house_num, i))
data = data_combiner.new_series(house_num, seed + i,
reconstruct_settings)
stats = realtime_disaggregator.run_trial(house_num, matcher_spec,
*data)
trial_stats.push(stats, 'a={},h={},i={}'.format(a, house_num, i))
return trial_stats
def plot_app_queries_func(stats: TrialStats):
questions_sets = stats.select('questions', lambda x: True)
all_labels = [chain(*[q.labels for q in qs]) for qs in questions_sets]
labels = collections.Counter(chain(*all_labels))
(labels, values) = map(
list,
zip(*sorted([(l, labels[l]) for l in labels], key=lambda x: x[0])))
print(f'There are {len(labels)} labels')
print(labels)
plt.xticks(range(len(values)), labels, rotation=45)
plt.bar(range(len(values)), values, label='Queries')
plt.ylabel('Queries')
def plot_app_votes_func(stats: TrialStats):
questions_sets = stats.select('questions', lambda x: True)
all_labels = chain(*[
chain(*[[(l, q.weight) for l in q.labels] for q in qs])
for qs in questions_sets
])
labels = collections.Counter()
for label, vote in all_labels:
labels[label] += vote
(labels, values) = map(
list,
zip(*sorted([(l, labels[l]) for l in labels], key=lambda x: x[0])))
plt.xticks(range(len(values)), labels, rotation=30)
plt.bar(range(len(values)), values, label='Votes')
plt.ylabel('Votes')
def plot_app_accuracies(stats: TrialStats):
mismatches = stats.select('mismatches', lambda t: True)
# Mismatches for last frame of each trial
last_mismatch = [m[-1] for m in mismatches]
failed_labels = []
for mismatch in last_mismatch:
failed_labels.extend(mismatch)
failures = collections.Counter(failed_labels)
# Remove always on from pairing
accuracies = sorted([(l, (len(mismatches) - failures[l]) / len(mismatches))
for l in failures if l != 'refrigerator'],
key=lambda x: x[0])
labels = sorted([l for l in failures])
print(f'There are {len(labels)} labels')
print(labels)
for i, (l, acc) in enumerate(accuracies):
plt.text(i, 1, f'{acc:.2f}', ha='center')
def plot_confidences(stats: TrialStats):
stats.plot('confidences', 'a=5')
stats.plot('confidences', 'a=7')
stats.plot('confidences', 'a=9')
plt.legend()
plt.show()
accuracies = sorted([(l, (len(mismatches) - failures[l]) / len(mismatches))
for l in failures if l != 'refrigerator'],
key=lambda x: x[0])
labels = sorted([l for l in failures])
print(f'There are {len(labels)} labels')
print(labels)
for i, (l, acc) in enumerate(accuracies):
plt.text(i, 1, f'{acc:.2f}', ha='center')
SEED = 434353456
path = Path(__file__).parent / f'{SEED}/accuracy_vs_time_vs_appliances'
stats = TrialStats.load(path)
# Aggregate over time
#plot_accuracies(stats)
#plot_queries(stats)
#plot_confidences(stats)
# Investigate: How often an appliance is 'on' vs how many times its queried
# Goal is to see if some appliances are disproportionatelly represented
# i.e. if an appliance is on for 1 hour, and another is on for 6
# the second should be queried 6x as much
# Maybe weight the first one since it's on 1/6th as much?
# Appliance breakdown
plot_app_queries_func(stats)
plot_app_votes_func(stats)
plot_app_accuracies(stats)