xdrl1/HeteroGDL-Indoor-Localization

The repo for the following paper: A Geometric Deep Learning Framework for Accurate Indoor Localization, 2022 International Conference on Indoor Positioning and Indoor Navigation (IPIN). IEEE, 2022.

HeteroGDL Indoor Localization IPIN2022

This is the document for the code used in the following paper:

Xuanshu Luo and Nirvana Meratnia. "A Geometric Deep Learning Framework for Accurate Indoor Localization" 2022 International Conference on Indoor Positioning and Indoor Navigation (IPIN). IEEE, 2022.

1. Environment Setup

The code is fully tested in CPU-only environment running Windows 11 Pro 22H2. The following steps should also be applicable in Linux environments.

1. Install miniconda. Please see Miniconda — Conda documentation

2. Setup the environment.

   conda env create -f env.yaml
   

   Note that the environment name is ipin2022 by default. You can edit the name field in the env.yaml file to specify the env name.

3. Activate the environment.

   conda activate ipin2022
   

   If you have changed the env name to your_env_name, then

   conda activate your_env_name
   

2. Obtaining Results

In this paper, two datasets are considered.

1. SoLoc: https://www.utwente.nl/en/eemcs/ps/dataset-folder/soloc-ipin2017-dataset.zip

   Paper: Le, Duc V., and Paul JM Havinga. "SoLoc: Self-organizing indoor localization for unstructured and dynamic environments." 2017 International Conference on Indoor Positioning and Indoor Navigation (IPIN). IEEE, 2017.

2. Petros: GitHub - pspachos/RSSI-Dataset-for-Indoor-Localization-Fingerprinting: RSSI dataset for Fingerprinting with Zigbee, BLE and WiFi

   Paper: S. Sadowski, P. Spachos, K. Plataniotis, "Memoryless Techniques and Wireless Technologies for Indoor Localization with the Internet of Things", IEEE Internet of Things Journal.

To run the code, please unzip the datasets above and specify the directories in _config.py

2.1 SoLoc

First go to the code directory for the SoLoc dataset.

cd ./code/SoLoc

Then load the dataset

python 0_load.py

2.1.1 WiFi only

First, please ensure that in ./SoLoc/_config.py, we keep the setting of WiFi only senario and comment out the setting for the other case

# WiFi only
EDGE_TYPES = ['bidirectWF']
N_AP = N_WIFI

# WiFi + Bluetooth
# EDGE_TYPES = ['bidirectWF', 'bidirectBT']
# N_AP = N_WIFI + N_BLUE

2.1.1.1 Data generation

In the WiFi only senario, you can generate two kinds of data, i.e., images and graphs using different command line options.

python 1_preprocess_wf.py i  (for images)
python 1_preprocess_wf.py g  (for graphs)

2.1.1.2 Apply different models

We can get seven results here. The w option means the scripts will consider the WiFi only data.

python 2_knn.py w                             (wkNN)
python 2_svm.py w                             (SVM)
python 2_mlp.py w                             (MLP)
python 2_cnn.py w                             (CNN)
python 2_hegcn.py --data w --aggr p           (GraphSAGE-pool)
python 2_hegcn.py --data w --aggr p --edge    (GraphSAGE-pool-edge)
python 2_hegcn.py --data w --aggr l --edge    (GraphSAGE-lstm-edge)

2.1.2 All together

First, please ensure that in ./SoLoc/_config.py, we keep the setting of WiFi+Bluetooth senario and comment out the setting for the WiFi only case.

# WiFi only
# EDGE_TYPES = ['bidirectWF']
# N_AP = N_WIFI

# WiFi + Bluetooth
EDGE_TYPES = ['bidirectWF', 'bidirectBT']
N_AP = N_WIFI + N_BLUE

2.1.2.1 Data generation

In this senario, you can generate three kinds of data, i.e., images and homogeneous graphs and heterogeneous graphs using different command line options.

python 1_preprocess_all.py i   (for images)
python 1_preprocess_all.py ho  (for homo. graphs)
python 1_preprocess_all.py he  (for hetero. graphs)

2.1.2.2 Apply different models

We can get eight results here. The a option means the scripts will consider data using WiFi and Bluetooth together.

python 2_knn.py a                             (wkNN)
python 2_svm.py a                             (SVM)
python 2_mlp.py a                             (MLP)
python 2_cnn.py a                             (CNN)
python 2_hogcn.py                             (HomoGraphSAGE-pool)
python 2_hegcn.py --data a --aggr p           (HeteroGraphSAGE-pool)
python 2_hegcn.py --data a --aggr p --edge    (HeteroGraphSAGE-pool-edge)
python 2_hegcn.py --data a --aggr l --edge    (HeteroGraphSAGE-lstm-edge)

2.2 Petros

First go to the code directory for the Petros dataset.

cd ./code/Petros

Note that we only consider the Scenario 1 and 3 in this project.

2.2.1 Scenario 1 (Bluetooth-only & All together)

To load the data in Scenario 1, for example, please run

python 0_load.py 1

2.2.1.1 Bluetooth only (Scenario 1)

First, please ensure that in ./Petros/_config.py, we keep the setting of Bluetooth only senario and comment out the setting for other cases.

# Bluetooth only
EDGE_TYPES = ['bidirectBT']
N_TOTAL_AP = N_BT

# WiFi only
# EDGE_TYPES = ['bidirectWF']
# N_TOTAL_AP = N_WF

# all together
# EDGE_TYPES = ['bidirectZB', 'bidirectBT', 'bidirectWF']
# N_TOTAL_AP = N_ZB + N_BT + N_WF

2.2.1.1.1 Data generation

In this senario, you can generate two kinds of data, i.e., images and graphs using different command line options.

python 1_preprocess_bt.py 1 i    (Scenario 1, images)
python 1_preprocess_bt.py 1 g    (Scenario 1, graphs)

2.2.1.1.2 Apply different models

We can get seven results here. The 1 and b option means the scripts will consider the Bluetooth only data in Scenario 1.

python 2_knn.py 1 b                          (wkNN)
python 2_svm.py 1 b                          (SVM)
python 2_mlp.py 1 b                          (MLP)
python 2_cnn.py 1 b                          (CNN)
python 2_hegcn.py 1 --data b --aggr p        (GraphSAGE-pool)
python 2_hegcn.py 1 --data b --aggr p --edge (GraphSAGE-pool-edge)
python 2_hegcn.py 1 --data b --aggr l --edge (GraphSAGE-lstm-edge)

2.2.1.2 All together (Scenario 1)

First, please ensure that in ./Petros/_config.py, we keep the setting of all three RF signals and comment out the setting for other cases.

# Bluetooth only
# EDGE_TYPES = ['bidirectBT']
# N_TOTAL_AP = N_BT

# WiFi only
# EDGE_TYPES = ['bidirectWF']
# N_TOTAL_AP = N_WF

# all together
EDGE_TYPES = ['bidirectZB', 'bidirectBT', 'bidirectWF']
N_TOTAL_AP = N_ZB + N_BT + N_WF

2.2.1.2.1 Data generation

In this senario, you can generate three kinds of data, i.e., images and homogeneous graphs and heterogeneous graphs using different command line options.

python 1_preprocess_all.py 1 i     (Scenario 1, images)
python 1_preprocess_all.py 1 ho    (Scenario 1, homo. graphs)
python 1_preprocess_all.py 1 he    (Scenario 1, hetero. graphs)

2.2.1.2.2 Apply different models

We can get eight results here. The 1 and a option means the scripts will consider the data using all kinds of RF in Scenario 1.

python 2_knn.py 1 a                          (wkNN)
python 2_svm.py 1 a                          (SVM)
python 2_mlp.py 1 a                          (MLP)
python 2_cnn.py 1 a                          (CNN)
python 2_hogcn.py 1                          (HomoGraphSAGE-pool)
python 2_hegcn.py 1 --data a --aggr p        (HeteroGraphSAGE-pool)
python 2_hegcn.py 1 --data a --aggr p --edge (HeteroGraphSAGE-pool-edge)
python 2_hegcn.py 1 --data a --aggr l --edge (HeteroGraphSAGE-lstm-edge)

2.2.2 Scenario 3 (WiFi-only & All together)

Similarly, for the Scenario 3, please run

python 0_load.py 3

Note that when generating new data, the existing data will be overwritten.

2.2.2.1 WiFi only (Scenario 3)

First, please ensure that in ./Petros/_config.py, we keep the setting of WiFi only senario and comment out the setting for other cases.

# Bluetooth only
# EDGE_TYPES = ['bidirectBT']
# N_TOTAL_AP = N_BT

# WiFi only
EDGE_TYPES = ['bidirectWF']
N_TOTAL_AP = N_WF

# all together
# EDGE_TYPES = ['bidirectZB', 'bidirectBT', 'bidirectWF']
# N_TOTAL_AP = N_ZB + N_BT + N_WF

2.2.2.1.1 Data generation

In this only senario, you can generate two kinds of data, i.e., images and graphs using different command line options.

python 1_preprocess_wf.py 3 i    (Scenario 3, images)
python 1_preprocess_wf.py 3 g    (Scenario 3, graphs)

2.2.2.1.2 Apply different models

We can get seven results here. The 3 and w option means the scripts will consider the WiFi only data in Scenario 3.

python 2_knn.py 3 w                          (wkNN)
python 2_svm.py 3 w                          (SVM)
python 2_mlp.py 3 w                          (MLP)
python 2_cnn.py 3 w                          (CNN)
python 2_hegcn.py 3 --data w --aggr p        (GraphSAGE-pool)
python 2_hegcn.py 3 --data w --aggr p --edge (GraphSAGE-pool-edge)
python 2_hegcn.py 3 --data w --aggr l --edge (GraphSAGE-lstm-edge)

2.2.2.2 All together (Scenario 3)

First, please ensure that in ./Petros/_config.py, we keep the setting of all three RF signals and comment out the setting for other cases.

# Bluetooth only
# EDGE_TYPES = ['bidirectBT']
# N_TOTAL_AP = N_BT

# WiFi only
# EDGE_TYPES = ['bidirectWF']
# N_TOTAL_AP = N_WF

# all together
EDGE_TYPES = ['bidirectZB', 'bidirectBT', 'bidirectWF']
N_TOTAL_AP = N_ZB + N_BT + N_WF

2.2.2.2.1 Data generation

In this senario, you can generate three kinds of data, i.e., images and homogeneous graphs and heterogeneous graphs using different command line options.

python 1_preprocess_all.py 3 i     (Scenario 3, images)
python 1_preprocess_all.py 3 ho    (Scenario 3, homo. graphs)
python 1_preprocess_all.py 3 he    (Scenario 3, hetero. graphs)

2.2.2.2.2 Apply different models

We can get eight results here. The 3 and a option means the scripts will consider the data using all kinds of RF in Scenario 3.

python 2_knn.py 3 a                          (wkNN)
python 2_svm.py 3 a                          (SVM)
python 2_mlp.py 3 a                          (MLP)
python 2_cnn.py 3 a                          (CNN)
python 2_hogcn.py 3                          (HomoGraphSAGE-pool)
python 2_hegcn.py 3 --data a --aggr p        (HeteroGraphSAGE-pool)
python 2_hegcn.py 3 --data a --aggr p --edge (HeteroGraphSAGE-pool-edge)
python 2_hegcn.py 3 --data a --aggr l --edge (HeteroGraphSAGE-lstm-edge)