- Run "util/save_transformed.py" with the dataset of interest uncommented.
- The aforementioned script may fail if the dataset relies on no digit 9 examples being present. If this is the case, use "util/remove_9s.ipynb" to modify the downloaded raw dataset before rerunning "util/data_synthesis/main.py".
usage: python main.py [-h] --input INPUT --model MODEL --experiment EXPERIMENT
example: python main.py --input colored_mnist --model resnet18 --experiment controlled_bias_training
All arguments are required and each must refer to an entry in its corresponding "config/" file's "options" dict.
- In "util/results_plotting.ipynb", fill the desired experiment parameters in and run
- Generate dataset for analysis using "util/save_dataset.ipynb"
- TODO(Jamell)
- Generate dataset for selectivity and invariance calculation using "util/invariance_data.ipynb"
- TODO(Jamell)
Use "util/results_plotting_loss_weight_tuning
Use the notebooks within "util/data_visualization"
Use "util/generalized combinations.ipynb"
Use "util/db.ipynb" to filter through and get timestamps and directories for experiments under some criteria