How to Run the Experiment

To run the experiment, execute the experiment script (model/run_compareExp.py). This script runs a series of tests with various combinations of model names, clustering methods, and dimensionality reduction methods.

1. Set up the Experiment:

• Import the Exp class from the modeling module.
• Initialize an instance of the Exp class.

2. Define Experiment Loops:

• The experiment will loop through multiple iterations (for i in range(10)), testing the models with different configurations.
• The models are tested for different seasons (summer, winter).
• Each model (ann_lin_dev, cnn_nn_dev, nn_lstm_nn_dev, DLinear, Autoformer, LSTNet) is tested under both clustered and non-clustered settings.

3. Running the Script:

• When executed, the script will automatically run all the configurations and generate results for each test.
• Output will include performance metrics for each model under the given conditions.

Parameters:

model_name

• Description: Specifies the model to test.
• Options:
  • ann_lin_dev: A type of Artificial Neural Network (ANN).
  • cnn_nn_dev: Convolutional Neural Network (CNN) based model.
  • nn_lstm_nn_dev: Neural Network with LSTM.
  • DLinear: A deep learning-based model.
  • Autoformer: A transformer-based model for time series forecasting.
  • LSTNet: A model combining LSTM with temporal convolution.

n_cluster

• Description: The number of clusters to use for clustering.
• Options:
  • 1: No clustering applied.
  • 2: Use 2 clusters.
  • 3: Use 3 clusters.
  • 4: Use 4 clusters.

cluster_method

• Description: The method to use for clustering.
• Options:
  • kmeans: K-means clustering.
  • None: No clustering method (i.e., single group).

season

• Description: The seasonality of the data.
• Options:
  • summer: Seasonality related to summer.
  • winter: Seasonality related to winter.

cluster_reduce_method

• Description: The method to use for reducing dimensions during clustering.
• Options:
  • pca: Principal Component Analysis, reduces dimensionality by projecting data onto the principal components.
  • tsne: t-Distributed Stochastic Neighbor Embedding, reduces dimensions while preserving local relationships.
  • umap: Uniform Manifold Approximation and Projection, preserves both local and global structure, faster than t-SNE.
  • None: No dimensionality reduction.

cluster_reduce_dim

• Description: The number of dimensions for the clustering reduction method.
• Options:
  • 2: Reduce to 2 dimensions.
  • None: No dimensionality reduction.