Pre-processing for Opportunity dataset - v2.0 release

[alessandro-montanari]

Hello,
I am trying to use the OpportunityDataset class (in data1.py) to parse the files in the dataset using the configuration below but it is not working. It cannot reshape the X tensor correctly at line 214
return np.reshape(res, (int(len(res) / win), win, self.config.f_num))

I was wondering how do you reduce the number of channels to 113 and which sensors are selected.
Thanks!

Config:

class OpportunityConfig(object):
    def __init__(self):
        ''' Without unlabeled data
        '''
        self.channels = 113
        self.f_num = self.channels
        self.overlap_ratio = 0.5
        self.s_win_size = 24
        self.c_win_size = 10  # 10
        self.s_labels_num = 4
        self.c_labels_num = 5
        self.batch_size = 64
        self.norm = False
        self.max_lr = 0.003
        self.min_lr = 0.0001
        self.decay_speed = 700
        self.iter = 2101
        self.test_point = 30
        self.dataset = data1.OpportunityDataset("OpportunityUCIDataset/dataset/", self)

[drewanye]

Hi, thanks for your question. We did not select all sensors in Opportunity dataset and selected the sensors which are important for activity recognition. I think the paper has described the sensors we adopted. Of course, you can define your own sensor channels depending on your situation and then modify the network. If you still have any problem, I am willing to revisit the work and give some detailed description. (Recently, I am a little busy. But I will try my best! Thanks.)

[Komodosis]

@alessandro-montanari Hello, may I ask you whether solve this question. I have also been troubled by this problem recently.