run_deepbiosim.sh

#!/usr/bin/env bash

INPUT_CSV="./input/gene_MGB_caries_transposed.csv" # Make sure each row represents a sample and each column represents a taxa/gene. Usually there are more taxa than samples. Index column and headers should be included. See example input file in ./input.
SIMULATION_METHOD="vae" # available simulation methods: vae, iwae, diffusion, kde. **kde may take forever to run** with high dimensional data with many samples.
MATCHING_METHOD="none" # available matching methods: rank (ordering sum of squares of features), mae (mean absolute error in bipartite matching), corr (Pearson correlation in bipartite matching). **rank is the fastest** and **corr uses Pearson correlation** is the most accurate.
OUTPUT_FOLDER="./output_sample/"

# **All parameters below are optional**

# ==== These parameters don't have to be the power of 2 ====
LATENT_DIM=16 # 8, 16, 32, 64 # as long as latent_dim <= number of samples
HIDDEN_DIM=128 # 64, 128, 256 as long as latent_dim <= hidden_dim
BATCH_SIZE=64 # 32, 64, 128
# ==========================================================

NUM_EPOCHS=100 # No need to adjust
LEARNING_RATE=0.001 # No need to adjust
K=10 # (for iwae only) the larger the more accurate but slower
TIME_STEPS=3000 # (for diffusion only) the larger the more accurate but slower
RANDOM_SEED=42 # reproducibility

# Run the simulation
for MATCHING_METHOD in "mae" #"rank"  "corr" "none"
do
  for SIMULATION_METHOD in "vae" "iwae" "diffusion"
  do
    python run_deepbiosim.py \
      --simulation_method $SIMULATION_METHOD \
      --matching_method $MATCHING_METHOD \
      --input $INPUT_CSV \
      --output_folder $OUTPUT_FOLDER \
      --latent_dim $LATENT_DIM \
      --hidden_dim $HIDDEN_DIM \
      --num_epochs $NUM_EPOCHS \
      --batch_size $BATCH_SIZE \
      --learning_rate $LEARNING_RATE \
      --K $K \
      --time_steps $TIME_STEPS \
      --random_seed $RANDOM_SEED
  done
done