The repository for AI4F services for updating building energy class and forecasting effect of potential benefit at installing solar panels.
This project is implemented in Docker providing a complete image for the entire service -- see Dockerfile and docker_compose.yml file for details regarding installation and requirements/dependencies
| Entrypoint | Filename |
|---|---|
| api | api.py |
| service 1 | classifier.py |
| service 2 | MLPRegresssor_HPO.py |
The project also includes:
- Dockerfile and docker_compose.yml: docker files responsible for deploying the respective image, as well as python_requirements.txt that contains the pip dependencies required to do so.
- service1_outputs.json and service1_outputs.json: JSON files that contain each product target’s name, description, data type. Paired with the respective model forecasts, they are utilized by API as format for the user response sent to front-end.
docker compose build
docker compose up -dOur pipelines are currently capable of processing a single csv file at a time, for the purposes of the service. Data must:
- be in csv format
- contain the columns registered either by the user-provided values or default values of the commnand-line arguments (see service 2 for more details)
Please fill command line arguments refering to columns of the dataset provided, otherwise the code will not function properly Services use python's click to edit parameters using command line arguments. Please refer to each service's documentation for further details.
Service 1 classification provides cross-validation grid search among many models and hyperparameter values to determine the most suitable combination:
| Algorithm | Parameters |
|---|---|
| KNN | {'n_neighbors': [1, 3, 5, 7, 9, 11, 13, 15], 'weights': ['uniform', 'distance']} |
| SVC | {'estimator__kernel': ['rbf', 'linear'], 'estimator__gamma': ['scale', 'auto'], 'estimator__C': [10, 100, 1000]} |
| LR | {'estimator__penalty': ['none', 'l2'], 'estimator__C': [1.0, 0.5, 0.1], 'estimator__solver': ['lbfgs', 'liblinear']} |
| Decision Tree | {'criterion': ['entropy'], 'max_depth': [6], 'min_samples_leaf': [1], 'min_samples_split': [4]} |
| Random Forest | {'n_estimators': [200, 600], 'max_depth': [4, 10, None], 'min_samples_leaf': [1, 2, 5]} |
| MLP | {'hidden_layer_sizes': [150, 100, 50], 'activation': ['relu', 'logistic', 'tanh'], 'solver': ['adam', 'lbfgs', 'sgd']} |
| XGBoost | {'learning_rate': [0.1, 0.2, 0.3], 'max_depth': [1, 2, 3, 4, 5, 6], 'min_child_weight': [1, 2], 'subsample': [1.0, 0.5, 0.1], 'n_estimators': [200, 600]} |
| Parameters | Type | Default Value | Description |
|---|---|---|---|
| input_filepath | str | './EF_comp.csv' | Folder path containing csv files used by the model |
| feature_cols | str | - | Dataset columns necesary for training |
| target_cols | str | - | Target column that we want to predict (model output) |
| output_dir | str | './models-scalers/' | local directory path to store models/scalers |
Example:
python classifier.py --input_filepath ./EF_comp.csv --feature_cols Building total area,Reference area,Above ground floors,Underground floor,Initial energy class,Energy consumption before,Energy class after
--target_cols Carrying out construction works,Reconstruction of engineering systems,Heat installation,Water heating system
--output_dir ./models-scalers/ Service 2 regression provides hyperparameter tuning on our MLP architecture to determine the right combination of hyperparameter values:
| Parameters | Type | Default Value | Description |
|---|---|---|---|
| input_filepath | str | './Sol_pan_comp.csv' | Folder path containing csv files used by the model |
| seed | str | '42' | seed used to set random state to the model |
| n_trials | int | '2' | number of trials - different tuning oh hyperparams |
| max_epochs | str | '3' | range of number of epochs used by the model |
| n_layers | str | '1' | range of number of layers used by the model |
| layer_sizes | str | "100" | range of size of each layer used by the model |
| activation | str | 'ReLU' | activation functions experimented on by the model |
| optimizer_name | str | 'Adam' | optimizers experimented on by the model |
| batch_size | str | '1024' | batch sizes experimented on by the model |
| n_trials | int | 50 | number of trials for HPO |
| num_workers | str | '2' | accelerator (cpu/gpu) processesors and threads used |
| preprocess | int | '1' | boolean if data require preprocessing and scaling |
| feature_cols | str | - | Dataset columns necesary for training |
| target_cols | str | - | Target column that we want to predict (model output) |
| output_dir | str | './models-scalers/' | local directory path to store models/scalers |
Example:
python MLPRegressor_HPO.py --dir_in ../Sol_pan_comp.csv/ --seed 42 --n_trials 20 --max_epochs 300 --n_layers 1 --layer_sizes 100 --l_window 240 --f_horizon 24
--l_rate 0.0001 --activation ReLU --optimizer_name Adam --batch_size 200 --needed_cols Region,Electricity consumption of the grid,Primary energy consumption before,Current inverter set power,Inverter power in project
--target_cols Electricity produced by solar panels --categorical_cols RegionFastAPI servers are the intermediate between the user interface and our model services, handling user requests, generating forecasts and providing the appropriate responses it uses swagger UI that, upon deployment, generates documentation that can be foun in localhost with port 8888 It deploys two (2) endpoints, one for each service with the following openAPI descriptions
| Description | Get building parameters service 1 |
|---|---|
| HTTP Method | POST |
| Endpoint URL | <host_ip>::8888/service_1/inference |
| Parameters | No parameters |
| Output example | [{"title":"Carrying out construction works","description":"Carrying out construction works in the enclosing structures during the project (to increase the energy efficiency of the house).","id":"1","value":"True"},{"title":"Reconstruction of engineering systems","description":"Reconstruction of engineering systems (ventilation, recuperation) to increase the energy efficiency of the house (during the project).","id":"2","value":"False"},{"title":"Water heating system","description":"Installation of a new water heating system (during the project).","id":"3","value":"False"},{"title":"Heat installation","description":"Installation of heat installations to ensure the production of heat from renewable energy sources.","id":"4","value":"False"}] |
| Example CURL request | curl -X 'POST' '<host_ip>:8888/service_1/inference' -H 'accept: application/json' -H 'Content-Type: application/json' -d '{"building_total_area": 351.6, "reference_area": 277.4, "above_ground_floors": 3, "underground_floor": 0, "initial_energy_class": "D", "energy_consumption_before": 106.04, "energy_class_after": "B"}' |
| Description | Get building parameters service 2 |
|---|---|
| HTTP Method | POST |
| Endpoint URL | <host_ip>::8888/service_2/inference |
| Parameters | No parameters |
| Output Example | [{"title":"Electricity produced by solar panels","description":"The amount of electricity produced by the solar panels, which are installed in the project.","id":"5","unit":"[MWh per year]","value":"7.45"},{"title":"Primary energy consumption after (KW)","description":"Primary energy consumption after installing the solar panel system.","id":"6","unit":"[MWh per year]","value":"0.45"},{"title":"Reduction of primary energy consumption","description":"Reduction of primary energy consumption: Difference between primary energy consumption before and after.","id":"7","unit":"[MWh per year]","value":"11.18"},{"title":"CO2 emissions reduction","description":"The amount of CO2 emissions reduction in the project.","id":"8","unit":"[tons of CO2 per year]","value":"0.81"}] |
| Example CURL request | curl -X 'POST' 'http://<host_ip>:8888/service_2/inference' -H 'accept: application/json' -H 'Content-Type: application/json' -d ' {"region": "Rīga", "electricity_consumption_of_the_grid": 4.65, "primary_energy_consumption_before": 11.63, "current_inverter_set_power": 0.0, "inverter_power_in_project": 10} |