forked from Rithwikksvr/EdgeNILM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
mini-experiments-test.py
238 lines (167 loc) · 8.74 KB
/
mini-experiments-test.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
import numpy as np
import pandas as pd
import time
import math
import matplotlib.pyplot as plt
from scipy.stats import pearsonr
import os
from sklearn.metrics import mean_absolute_error, f1_score
import math
import torch
from torch import nn
import torch.nn.functional as F
import torch.optim as optim
from collections import OrderedDict
import time
import sys
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
from functions import *
from networks import Seq2Point
def test_fold(model_name, appliances, fold_number, sequence_length, batch_size, results_arr):
dir_name = "fold_%s_models"%(fold_number)
dir_name = os.path.join(dir_name, "sequence_length_%s"%(sequence_length))
dir_name = os.path.join(dir_name, model_name)
models_dir = dir_name
test_file_name = 'test_%s.h5'%(fold_number)
all_appliances_mains_lst, all_appliances_truth = load_h5_file(test_file_name, appliances)
# Taking the first 10% or 20%
for i in range(len(all_appliances_mains_lst)):
n = len(all_appliances_mains_lst[i])
for j in range(n):
df = all_appliances_mains_lst[i][j]
all_appliances_mains_lst[i][j] = df.iloc[:int(fraction_to_test * len(df))]
df = all_appliances_truth[i][j]
all_appliances_truth[i][j] = df.iloc[:int(fraction_to_test * len(df))]
if method!='zero':
parameters_path = os.path.join(dir_name, 'parameters.json')
f = open(parameters_path)
parameters = json.load(f)
all_appliances_predictions = []
for appliance_index, appliance_name in enumerate(appliances):
mains_mean = parameters[appliance_name]['mains_mean']
mains_std = parameters[appliance_name]['mains_std']
app_mean = parameters[appliance_name]['app_mean']
app_std = parameters[appliance_name]['app_std']
appliance_mains_dfs = all_appliances_mains_lst[appliance_index]
no_of_homes = len(appliance_mains_dfs)
if 'mtl' not in model_name:
model_path = os.path.join(dir_name, "%s.pth"%(appliance_name))
else:
model_path = os.path.join(dir_name, "weights.pth")
if not cuda:
model = torch.load(model_path,map_location=torch.device('cpu'))
else:
model = torch.load(model_path)
model.eval()
appliance_prediction = []
for home_id in range(no_of_homes):
home_mains = appliance_mains_dfs[home_id]
l = len(home_mains)
processed_mains = mains_preprocessing([home_mains], sequence_length)
processed_mains = (processed_mains - mains_mean)/mains_std
if 'mtl' in model_name:
prediction = predict_mtl(model, processed_mains, appliance_index, cuda, batch_size)
else:
prediction = predict(model, processed_mains, cuda, batch_size)
prediction = prediction * app_std + app_mean
prediction = prediction.flatten()
prediction = np.where(prediction>0, prediction,0)
df = pd.DataFrame({appliance_name: prediction})
df.index = home_mains.index
appliance_prediction.append(df)
# print (home_mains.shape)
# print ()
all_appliances_predictions.append(appliance_prediction)
# print ("Finished predicting for appliance %s"%(appliance_name))
else:
all_appliances_predictions = []
for appliance_index, appliance_name in enumerate(appliances):
appliance_mains_dfs = all_appliances_mains_lst[appliance_index]
no_of_homes = len(appliance_mains_dfs)
appliance_prediction = []
for home_id in range(no_of_homes):
home_mains = appliance_mains_dfs[home_id]
l = len(home_mains)
prediction = np.zeros(l)
df = pd.DataFrame({appliance_name: prediction})
df.index = home_mains.index
appliance_prediction.append(df)
# print (home_mains.shape)
# print ()
all_appliances_predictions.append(appliance_prediction)
results = []
results.append(model_name)
results.append(sequence_length)
results.append(fold_number)
results.append(batch_size)
total_error = [0 for q in range(len(metrics))]
for app_index, app_name in enumerate(appliances):
for metric_index, metric in enumerate(metrics):
if metric=='mae':
truth_ = pd.concat(all_appliances_truth[app_index],axis=0).values
pred_ = pd.concat(all_appliances_predictions[app_index],axis=0).values
error = mean_absolute_error(truth_, pred_)
elif metric=='f1-score':
truth_ = pd.concat(all_appliances_truth[app_index],axis=0).values
pred_ = pd.concat(all_appliances_predictions[app_index],axis=0).values
truth_ = np.where(truth_>threshold, 1,0)
pred_ = np.where(pred_>threshold, 1,0)
error = f1_score(truth_, pred_)
else:
total_ground_truth_usage = [np.sum(df.values) for df in all_appliances_truth[app_index]]
total_prediction_usage = [np.sum(df.values) for df in all_appliances_predictions[app_index]]
total_mains_usage = [np.sum(df.values) for df in all_appliances_mains_lst[app_index]]
energy_incorrectly_assigned = [ (total_ground_truth_usage[home]/total_mains_usage[home] )- (total_prediction_usage[home]/total_mains_usage[home]) for home in range(len(total_ground_truth_usage)) ]
error = 100*np.mean(np.abs(energy_incorrectly_assigned))
print ("%s %s Error: %s"%(app_name, metric,error))
results.append(error)
total_error[metric_index]+=error
if plot:
truth_ = pd.concat(all_appliances_truth[app_index],axis=0).values
pred_ = pd.concat(all_appliances_predictions[app_index],axis=0).values
plt.figure(figsize=(30,4))
plt.plot(truth_[:1000],'r',label="Truth")
plt.plot(pred_[:1000],'b',label="Pred")
plt.legend()
plt.savefig("images/%s_%s_%s_fold_%s.png"%(model_name, app_name,sequence_length,fold_number))
plt.close()
if save_predictions:
truth_ = pd.concat(all_appliances_truth[app_index],axis=0).values
pred_ = pd.concat(all_appliances_predictions[app_index],axis=0).values
np.save("predictions/%s_fold_%s.png"%( app_name,fold_number),truth_)
np.save("predictions/%s_%s_%s_fold_%s.png"%(model_name, app_name,sequence_length,fold_number),pred_)
results = results + total_error
results_arr.append(results)
return all_appliances_truth, all_appliances_predictions
appliances = ["fridge",'dish washer','washing machine']
appliances.sort()
batch_size=4096
fold_numbers=[1, 2, 3]
sequence_lengths = [99]
fraction_to_test = 1
cuda=True
plot=False
save_predictions=True
metrics = ['mae','f1-score','sae']
threshold = 15
create_dir_if_not_exists('mini-experiments-results')
for method in ['only_convolutions_pruned_model_30_percent','only_neurons_pruned_model_30_percent','not_fully_shared_mtl']:
results_arr = []
for fold_number in fold_numbers:
print ("Batch size:", batch_size)
for sequence_length in sequence_lengths:
print ("-"*50)
print ("Results %s; sequence length: %s "%(method, sequence_length))
truth, all_predictions = test_fold(method, appliances, fold_number, sequence_length, batch_size, results_arr)
print ("-"*50)
print ("\n\n\n")
columns = ['Model Name',"Sequence Length","Fold Number","Batch Size"]
for app_name in appliances:
for metric in metrics:
columns.append(metric+ app_name+" Error")
for metric in metrics:
columns.append("Total "+metric)
results_arr= np.array(results_arr)
df = pd.DataFrame(data=results_arr, columns=columns, index = range(len(results_arr)))
df.to_csv(os.path.join('mini-experiments-results','%s.csv'%(method)),index=False)