def get_best_instance_model(instance_models, device_type, key_for_model_name):
dfs_model = {}
best_model_score = 0
for model_name in instance_models:
instances_of_model = []
for instance in device_type.instances:
test_trace = instance.traces[0]
instance_name = test_trace.metadata[key_for_model_name]
test = utils.trace_series_to_numpy_array(test_trace.series)
model_score = instance_models[model_name].score(test)
instances_of_model.append([model_name, instance_name, model_score])
if (model_score > best_model_score):
best_model = instance_models[model_name]
dfs_model[model_name] = pd.DataFrame(
data=instances_of_model,
columns=['Model_Instance', 'Test_Instance', 'Value'])
model_averages = []
for key in dfs_model:
sum = 0
for row in dfs_model[key].iterrows():
sum = sum + row[1]['Value']
model_averages.append([key, sum / len(dfs_model[key].index)])
print
avg_model_df = pd.DataFrame(data=model_averages,
columns=['Model_Instance', 'Avg Probability'])
print avg_model_df._sort('Avg Probability', ascending=False)
bestModel = avg_model_df._sort('Avg Probability', ascending=False)._sort(
'Avg Probability', ascending=False).head(1)['Model_Instance'].values[0]
print str(bestModel) + ' is best.'
return bestModel
def get_best_instance_model(instance_models,device_type,key_for_model_name):
dfs_model = {}
best_model_score = 0
for model_name in instance_models:
instances_of_model = []
for instance in device_type.instances:
test_trace = instance.traces[0]
instance_name = test_trace.metadata[key_for_model_name]
test = utils.trace_series_to_numpy_array(test_trace.series)
model_score = instance_models[model_name].score(test)
instances_of_model.append([model_name,instance_name,model_score])
if(model_score > best_model_score):
best_model = instance_models[model_name]
dfs_model[model_name] = pd.DataFrame(data=instances_of_model,
columns=['Model_Instance','Test_Instance','Value'])
model_averages = []
for key in dfs_model:
sum=0
for row in dfs_model[key].iterrows():
sum = sum+row[1]['Value']
model_averages.append([key,sum/len(dfs_model[key].index)])
print
avg_model_df = pd.DataFrame(data=model_averages,
columns=['Model_Instance','Avg Probability'])
print avg_model_df._sort('Avg Probability',ascending=False)
bestModel = avg_model_df._sort('Avg Probability',
ascending=False)._sort('Avg Probability',
ascending=False).head(1)['Model_Instance'].values[0]
print str(bestModel) + ' is best.'
return bestModel
def fit_trace_to_HMM(model, trace):
'''
Fits the given trace to the model. NaNs are turned into zeroes.
'''
trace_values = utils.trace_series_to_numpy_array(trace.series)
model.fit([trace_values])
startprob, means, covars, transmat = _sort_learnt_parameters(
model.startprob_, model.means_, model.covars_, model.transmat_)
model = hmm.GaussianHMM(startprob.size, 'full', startprob, transmat)
model.means_ = means
model.covars_ = covars
return model
def fit_trace_to_HMM(model,trace):
'''
Fits the given trace to the model. NaNs are turned into zeroes.
'''
trace_values = utils.trace_series_to_numpy_array(trace.series)
model.fit([trace_values])
startprob, means, covars, transmat = _sort_learnt_parameters(model.startprob_,
model.means_, model.covars_ , model.transmat_)
model=hmm.GaussianHMM(startprob.size, 'full', startprob, transmat)
model.means_ = means
model.covars_ = covars
return model
def disaggregate_data(model_tuple, trace):
data=[]
power_total=utils.trace_series_to_numpy_array(trace.series)
[decoded_states, decoded_power]=predict_with_FHMM(model_tuple[0],
model_tuple[1],model_tuple[2],power_total)
for i,v in enumerate(decoded_power['air1']):
date_time=trace.series.index[i]
value=trace.series[i]
data.append({'date':date_time.strftime('%Y-%m-%d %H:%M'),
'dg': float(v),'reading':float(value)})
json_string = json.dumps(data, ensure_ascii=False,indent=4,
separators=(',', ': '))
return json_string