Added leave-1 and leave-n out imports

Model Development and Deployment/TinyML Models/ProtoNN/data_utils.py

@@ -60,5 +60,182 @@ def import_auritus_activity_dataset(dataset_folder = 'Train_Val_test/', use_time
     else:
         return X_tr, Y_tr
 
+def import_auritus_activity_dataset_one_out(dataset_folder = 'Train_Val_test/', use_timestamp=False, shuffle=True, window_size = 550, stride = 50,channels=0):
+    if(use_timestamp==True and channels==0):
+        X_tr = np.empty([0, window_size, 7])
+        X_test = np.empty([0, window_size, 7])
+    elif(use_timestamp==False and channels==2):
+        X_tr = np.empty([0, window_size, 2])
+        X_test = np.empty([0, window_size, 2])
+    elif(use_timestamp==True and channels==2):
+        X_tr = np.empty([0, window_size, 3])
+        X_test = np.empty([0, window_size, 3])
+    else:
+        X_tr = np.empty([0, window_size, 6])
+        X_test = np.empty([0, window_size, 6])
+    Y_tr = np.empty([0,1])
+    Y_test = np.empty([0,1])
+    train_file_list = os.listdir(dataset_folder)
+    if('.DS_Store' in train_file_list):
+        train_file_list.remove('.DS_Store')
+
+    one_out_num = random.randint(1,45)
+    print('One out participant: ',one_out_num)
+    tf_list_edited = []
+    test_list_edited = []
+    for i in range(len(train_file_list)):
+        if int(re.findall(r'\d+', train_file_list[i])[0]) != one_out_num:
+            tf_list_edited.append(train_file_list[i])
+        else:
+            test_list_edited.append(train_file_list[i])
+
+    labels = ['W','R','J','St','Tl','Tr','Si','L','F']
+    for line in tqdm(tf_list_edited):
+        if(use_timestamp==True):
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5,6]).to_numpy()
+        else:
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5]).to_numpy()
+        if(channels==2):
+            acc = np.sqrt(cur_train[:,0]*cur_train[:,0] + cur_train[:,1]*cur_train[:,1] +cur_train[:,2]*cur_train[:,2])
+            gyr = np.sqrt(cur_train[:,3]*cur_train[:,3] + cur_train[:,4]*cur_train[:,4] +cur_train[:,5]*cur_train[:,5])
+            if(use_timestamp==False):
+                cur_train = np.transpose(np.vstack((acc,gyr)))
+            else:
+                time_vec = cur_train[:,6]
+                cur_train = np.transpose(np.vstack((acc,gyr,time_vec)))
+        cur_label = labels.index([ele for ele in labels if(ele in line)][0])
+        windows = SlidingWindow(size=window_size, stride=stride)
+        cur_train_3D = windows.fit_transform(cur_train[:,0])
+        for i in range(1,cur_train.shape[1]):
+            X_windows = windows.fit_transform(cur_train[:,i])
+            cur_train_3D = np.dstack((cur_train_3D,X_windows))  
+        cur_GT = cur_label * np.ones((cur_train_3D.shape[0],1))
+        X_tr = np.vstack((X_tr, cur_train_3D))
+        Y_tr = np.concatenate((Y_tr, cur_GT))
+    Y_tr = Y_tr.flatten()
+    if(shuffle==True):
+        shuffler = np.random.permutation(X_tr.shape[0])
+        X_tr = X_tr[shuffler]
+        Y_tr = Y_tr[shuffler]
+        Y_tr = to_categorical(Y_tr,9)
+
+    for line in tqdm(test_list_edited):
+        if(use_timestamp==True):
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5,6]).to_numpy()
+        else:
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5]).to_numpy()
+        if(channels==2):
+            acc = np.sqrt(cur_train[:,0]*cur_train[:,0] + cur_train[:,1]*cur_train[:,1] +cur_train[:,2]*cur_train[:,2])
+            gyr = np.sqrt(cur_train[:,3]*cur_train[:,3] + cur_train[:,4]*cur_train[:,4] +cur_train[:,5]*cur_train[:,5])
+            if(use_timestamp==False):
+                cur_train = np.transpose(np.vstack((acc,gyr)))
+            else:
+                time_vec = cur_train[:,6]
+                cur_train = np.transpose(np.vstack((acc,gyr,time_vec)))
+        cur_label = labels.index([ele for ele in labels if(ele in line)][0])
+        windows = SlidingWindow(size=window_size, stride=stride)
+        cur_train_3D = windows.fit_transform(cur_train[:,0])
+        for i in range(1,cur_train.shape[1]):
+            X_windows = windows.fit_transform(cur_train[:,i])
+            cur_train_3D = np.dstack((cur_train_3D,X_windows))  
+        cur_GT = cur_label * np.ones((cur_train_3D.shape[0],1))
+        X_test = np.vstack((X_test, cur_train_3D))
+        Y_test = np.concatenate((Y_test, cur_GT))
+    Y_test = Y_test.flatten()
+    if(shuffle==True):
+        shuffler = np.random.permutation(X_test.shape[0])
+        X_test = X_test[shuffler]
+        Y_test = Y_test[shuffler]
+        Y_test = to_categorical(Y_test,9)
+
+    return X_tr, Y_tr, X_test, Y_test, one_out_num
 
 
+def import_auritus_activity_dataset_multi_out(dataset_folder = 'Train_Val_test/', use_timestamp=False, shuffle=True, window_size = 550, stride = 50,channels=0, num_out = 10):
+    if(use_timestamp==True and channels==0):
+        X_tr = np.empty([0, window_size, 7])
+        X_test = np.empty([0, window_size, 7])
+    elif(use_timestamp==False and channels==2):
+        X_tr = np.empty([0, window_size, 2])
+        X_test = np.empty([0, window_size, 2])
+    elif(use_timestamp==True and channels==2):
+        X_tr = np.empty([0, window_size, 3])
+        X_test = np.empty([0, window_size, 3])
+    else:
+        X_tr = np.empty([0, window_size, 6])
+        X_test = np.empty([0, window_size, 6])
+    Y_tr = np.empty([0,1])
+    Y_test = np.empty([0,1])
+    train_file_list = os.listdir(dataset_folder)
+    if('.DS_Store' in train_file_list):
+        train_file_list.remove('.DS_Store')
+
+    multi_out_num = random.sample(range(0, 45), num_out)
+    print('Multi out participant: ',multi_out_num)
+    tf_list_edited = []
+    test_list_edited = []
+    for i in range(len(train_file_list)):
+        if int(re.findall(r'\d+', train_file_list[i])[0]) in multi_out_num: 
+            test_list_edited.append(train_file_list[i])
+        else:
+            tf_list_edited.append(train_file_list[i])
+
+    labels = ['W','R','J','St','Tl','Tr','Si','L','F']
+    for line in tqdm(tf_list_edited):
+        if(use_timestamp==True):
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5,6]).to_numpy()
+        else:
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5]).to_numpy()
+        if(channels==2):
+            acc = np.sqrt(cur_train[:,0]*cur_train[:,0] + cur_train[:,1]*cur_train[:,1] +cur_train[:,2]*cur_train[:,2])
+            gyr = np.sqrt(cur_train[:,3]*cur_train[:,3] + cur_train[:,4]*cur_train[:,4] +cur_train[:,5]*cur_train[:,5])
+            if(use_timestamp==False):
+                cur_train = np.transpose(np.vstack((acc,gyr)))
+            else:
+                time_vec = cur_train[:,6]
+                cur_train = np.transpose(np.vstack((acc,gyr,time_vec)))
+        cur_label = labels.index([ele for ele in labels if(ele in line)][0])
+        windows = SlidingWindow(size=window_size, stride=stride)
+        cur_train_3D = windows.fit_transform(cur_train[:,0])
+        for i in range(1,cur_train.shape[1]):
+            X_windows = windows.fit_transform(cur_train[:,i])
+            cur_train_3D = np.dstack((cur_train_3D,X_windows))  
+        cur_GT = cur_label * np.ones((cur_train_3D.shape[0],1))
+        X_tr = np.vstack((X_tr, cur_train_3D))
+        Y_tr = np.concatenate((Y_tr, cur_GT))
+    Y_tr = Y_tr.flatten()
+    if(shuffle==True):
+        shuffler = np.random.permutation(X_tr.shape[0])
+        X_tr = X_tr[shuffler]
+        Y_tr = Y_tr[shuffler]
+        Y_tr = to_categorical(Y_tr,9)
+
+    for line in tqdm(test_list_edited):
+        if(use_timestamp==True):
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5,6]).to_numpy()
+        else:
+            cur_train = pd.read_csv(dataset_folder+line,header=None,usecols=[0,1,2,3,4,5]).to_numpy()
+        if(channels==2):
+            acc = np.sqrt(cur_train[:,0]*cur_train[:,0] + cur_train[:,1]*cur_train[:,1] +cur_train[:,2]*cur_train[:,2])
+            gyr = np.sqrt(cur_train[:,3]*cur_train[:,3] + cur_train[:,4]*cur_train[:,4] +cur_train[:,5]*cur_train[:,5])
+            if(use_timestamp==False):
+                cur_train = np.transpose(np.vstack((acc,gyr)))
+            else:
+                time_vec = cur_train[:,6]
+                cur_train = np.transpose(np.vstack((acc,gyr,time_vec)))
+        cur_label = labels.index([ele for ele in labels if(ele in line)][0])
+        windows = SlidingWindow(size=window_size, stride=stride)
+        cur_train_3D = windows.fit_transform(cur_train[:,0])
+        for i in range(1,cur_train.shape[1]):
+            X_windows = windows.fit_transform(cur_train[:,i])
+            cur_train_3D = np.dstack((cur_train_3D,X_windows))  
+        cur_GT = cur_label * np.ones((cur_train_3D.shape[0],1))
+        X_test = np.vstack((X_test, cur_train_3D))
+        Y_test = np.concatenate((Y_test, cur_GT))
+    Y_test = Y_test.flatten()
+    if(shuffle==True):
+        shuffler = np.random.permutation(X_test.shape[0])
+        X_test = X_test[shuffler]
+        Y_test = Y_test[shuffler]
+        Y_test = to_categorical(Y_test,9)
+