data_loader.py

import os
import shutil
import zipfile

import numpy as np
from absl import app, flags
from scipy import stats
from scipy.fftpack import fft
from scipy.io import loadmat
from sklearn import utils as skutils

class UNIMIB( object ):

    def __init__( self ):
        self._path              = '/data/datasets/UNIMIBDataset/'
        self._name              = 'unimib'
        self._channel           = 3
        self._winlen            = 100
        self._user_num          = 30
        self._act_num           = 9
        self._train_user_num    = 30
        self._train_act_num     = 9
        self._data_shape        = [ None, self._winlen, self._channel, 1 ]

    def load_data( self, step = 0 ):

        TrainData   = np.empty( [0, self._winlen, self._channel], dtype=np.float )
        TrainLA     = np.empty( [0], dtype=np.int32 )
        TrainLU     = np.empty( [0], dtype=np.int32 )
        TrainID     = np.empty( [0], dtype=np.int32 )

        TestData   = np.empty( [0, self._winlen, self._channel], dtype=np.float )
        TestLA     = np.empty( [0], dtype=np.int32 )
        TestLU     = np.empty( [0], dtype=np.int32 )
        TestID     = np.empty( [0], dtype=np.int32 )

        for i in range(10):
            Data    = np.load( self._path + 'OuterPartition/fold{}/data.npy'.format(i) )
            LA      = np.load( self._path + 'OuterPartition/fold{}/act_label.npy'.format(i) )
            LU      = np.load( self._path + 'OuterPartition/fold{}/user_label.npy'.format(i) )
            ID      = np.load( self._path + 'OuterPartition/fold{}/id.npy'.format(i) )

            if step == i:
                TestData   = np.concatenate( (TestData, Data), axis=0 )
                TestLA     = np.concatenate( (TestLA, LA), axis=0 )
                TestLU     = np.concatenate( (TestLU, LU), axis=0 )
                TestID     = np.concatenate( (TestID, ID), axis=0 )
            else:
                TrainData   = np.concatenate( (TrainData, Data), axis=0 )
                TrainLA     = np.concatenate( (TrainLA, LA), axis=0 )
                TrainLU     = np.concatenate( (TrainLU, LU), axis=0 )
                TrainID     = np.concatenate( (TrainID, ID), axis=0 )

        # delete overlap samples form training_data
        overlap_ID  = np.unique( np.concatenate( ( TestID+1, TestID-1) ) )
        invalid_idx = np.array( [ i for i in np.arange( len(TrainID) ) if TrainID[i] in overlap_ID ] )

        TrainData  = np.delete( TrainData, invalid_idx, axis=0 )
        TrainLA    = np.delete( TrainLA,   invalid_idx, axis=0 )
        TrainLU    = np.delete( TrainLU,   invalid_idx, axis=0 )
        # deleted

        # normalization
        mean    = np.mean( np.reshape(  TrainData, [-1, self._channel] ), axis=0 )
        std     = np.std( np.reshape(   TrainData, [-1, self._channel] ), axis=0 )

        TrainData  = (TrainData - mean)/std
        TestData   = (TestData - mean)/std

        TrainData  = np.expand_dims( TrainData, 3 )
        TestData   = np.expand_dims( TestData,  3 )

        return TrainData, TrainLA, TrainLU, TestData, TestLA, TestLU