Setup

Requirements

• Python 3.9.10
• Packages and package versions as in requirements.txt

Virtual Environment

Run the following commands in the repository root to create a virtual environment with all dependencies:

python -m venv setml-env
source setml-env/bin/activate
python -m pip install -r requirements.txt

Cloning Model Code

From the repository root, move to the setml folder and clone the Set Transformer repository:

git clone https://github.com/juho-lee/set_transformer.git

as well as the FSPool repository:

git clone https://github.com/Cyanogenoid/fspool.git

Running Experiments

To set up an experiment, change the config/config.yaml file to the desired experimental setup, then in the root directory of this repository run:

python main.py

Desperate Student M-Tuple

To run models optimized for the Desperate Student M-Tuple task, use the following commands:

Deep Sets

python main.py model.type=deepsets_ds1t model.latent_dim=5

Sorted MLP

python main.py model.type=sorted_mlp_ds1t model.latent_dim=5

Set Transformer

python main.py model.type=set_transformer_ds1t model.latent_dim=64 experiment.batch_size=16 experiment.use_batch_sampler=True

Available Models

Model	model_type
DeepSets with MLP	deepsets_mlp
Principal Neighbourhood Aggregation	pna
Small Set Transformer	small_set_transformer
MLP	mlp
Sorted MLP	sorted_mlp
DeepSets with MLP and FSPool	deepsets_mlp_fspool

Available Labels (Tasks)

Name	label	Description	Set→Label	Multiset→Label
Sum	sum	Sum of all elements in set	{4, 2, 1}→7	{4, 2, 1, 1}→8
Mean	mean	Mean of all elements in set	{4, 2, 1}→2.333..	{4, 2, 1, 1}→2
Mode	mode	Mode value in the Set	NA	{4, 2, 1, 1}→1
Maximum	max	Maximum value in the Set	{4, 2, 1}→4	{4, 2, 1, 1}→4
Cardinality	cardinality	Cardinality (size) of the set	{4, 2, 1}→3	{4, 2, 1, 1}→4
Longest Sequence Length	longest_seq_length	Length of the longest monotonically increasing sequence	{2, 1, 4}→2	{2, 1, 1, 4}→3
Longest Contiguous Sum	largest_contiguous_sum	Largest sum of the subset of the set	{2, 1, 4, -10, 7}→14	{2, 1, 1, 4, -10}→8
Largest pair sum	largest_pair_sum	Largest sum of a pair of numbers in the set	{2, 1, 4, -10, 7}→11	{2, 1, 1, 4, -10}→6
Largest triplet sum	largest_triple_sum	Largest sum of a triplet of numbers in the set	{2, 1, 4, -10, 7}→13	{2, 1, 1, 4, -10}→7
Contains even number	contains_even	1 if the set contains an even number, 0 otherwise	{2, 1, 4, -10, 7}→1	{3, 1, 1, 5, -9}→0

For the Desperate Student M-Tuple task, use:

desperate_student_<M>_tuple

and replace <M> by the desired order of the tuple, available are 1 to 6 so for example use for triples use:

desperate_student_3_tuple

Unit Tests

We use pytest for unit testing. To run the unit tests, execute the following command in the root directory of this repository:

pytest --ignore=setml/fspool