This repository is the official implementation of Cohort comfort models - Using occupants' similarity to predict personal thermal preference with less data.
conda env create --file envs/environment_<OS>.yaml # replace OS with either `macos` or `ubuntu`
We chose the following publicly available field longitudinal datasets:
Once the datasets are downloaded into data/, the notebook preprocessing.ipynb does some pre-processing and creates copies of the datasets with different feature sets.
Each longitudinal dataset has primarily two files:
- One with sensor measurements with the right-here-right-now (RHRH) responses for all participants.
- Another one with the on-boarding one-time survey responses for each participant.
All the pre-processing is handle by the class DataLoader defined in data.py.
The files needed by the preprocessing.ipynb notebook are:
| File name | Features |
|---|---|
df_info_subjects_raw_DORN.csv |
Original DORN dataset with occupants' information |
df_cozie_env_raw_DORN.csv |
Original DORN dataset with environmental and RHRN labels |
210316_SMC_background.csv |
Original SMC dataset with occupants' information |
SMC_combined_survey2.csv |
Original SMC dataset with environmental and RHRN labels |
All the needed files with the selected features sets, onboarding questions, and number of datapoints per person are defined inside the notebook.
To specify different settings and parameters pertinent to the data being loaded, training, and evaluation, modify the files inside the configs/ folder as a yaml file.
Inside, a description of existing config files resides in config_details.xslx.
Personal comfort models (PCM), general-purpose, and cohort models are trained using Cross-Validation (CV) described in utils.py under the train_model function. The chosen model for all approaches is Random Decision Forest (RDF).
In order to create the different cohorts and assign occupants to them, user profiling is done via the class defined in user_profile.py. This class takes care of the cross-model performance between occupants PCMs,the Jensen-Shannon divergence between their distribution responses, and the onboarding responses clustering.
The file cohort.py handles the cohorts themselves
Finally, the file cohort_modeling.py is an end-to-end initialisation, training, and evaluation of the different cohort modeling approaches. By default the entire process is repeated 100 times to account for stochasticity.
./run_cohort_model.sh configs/{your_config}.yaml
The configuration files used for manuscript can be found in the foulder configs/ alongside a spreadsheet config_details.xlsx with details about each configuration file.
Trained personal comfort models can be found in the models/ folder. They are saved in pickle format as dictionaries.
It is recommended to have the PCM already computed to minimise computation time. The file names are as follows:
dict_pcm_{model}_{dataset_type}_{dataset_name}_{metric}.pickle
where,
{model}is the classification model used, in this work was RDF.{dataset_type}is the feature set used for training the model,sensorrefer to environmental features and physiological measurements{dataset_name}is one, or the combination, of the public dataset used.{metric}is the performance metric used, eitherf1_microorcohen_kappa
To calculate the PCM according to the different settings run the following:
python train_pcm.py configs/train_pcm_{dataset}.yaml
Example: python train_pcm.py configs/train_pcm_dorn.yaml
The configuration files used for manuscript can be found in the foulder configs/ alongside a spreadsheet config_details.xlsx with details about each configuration file.
All results files are generated under experiments_{DATASET} on a folder titled after the name parameter in the respective configuration file used.
Example: ./run_cohort_model.sh configs/experiment_20210407.yaml uses the DORN dataset, thus the results files will be saved under experiments_dorn/20210407/
All images used in the manuscript can be generated by following the jupyter notebook final-plots.ipynb and are stored under the folder img/
The notebook user_exploration.ipynb provides the analysis for the occupant's metadata breakdown beased on the cohort approach used (Table Appendix A.2 in the manucript)
MIT License
Copyright (c) 2022 Building and Urban Data Science (BUDS) Group and National University of Singapore (NUS)
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.