run_image_exp.sh

#!/bin/bash

# Define datasets
DATASETS=("CIFAR-10" "CIFAR-100" "SVHN" "FashionMNIST")

# Define models
MODELS=("vit_base.yaml" "conv_base.yaml")

# Configuration
SEED=20
BASE_PORT=29500

# Create logs directory
mkdir -p logs

echo "Starting image dataset experiments..."
echo "Datasets: ${DATASETS[*]}"
echo "Models: ${MODELS[*]}"
echo "Seed: $SEED"
echo "========================================"

for DATASET in "${DATASETS[@]}"; do
    echo "Processing dataset: $DATASET"
    
    # Ensure logs directory for dataset exists
    mkdir -p "logs/${DATASET}"
    
    PIDS=()
    
    for MODEL_CFG in "${MODELS[@]}"; do
        MODEL_NAME=$(basename "$MODEL_CFG" .yaml)
        CFG="cfgs/${DATASET}/${MODEL_CFG}"
        
        # Check if config exists
        if [ ! -f "$CFG" ]; then
            echo "  Warning: Config $CFG not found, skipping..."
            continue
        fi
        
        # Use simple tag like command.txt
        TAG="seed_${SEED}"
        
        # Unique port for each run to be safe
        PORT=$BASE_PORT
        BASE_PORT=$((BASE_PORT + 1))
        
        echo "  Running $MODEL_NAME on $DATASET (Port: $PORT)..."
        
        # Run command in background
        # Using nproc_per_node=4 as these are image models (referencing command.txt)
        nohup torchrun --nproc_per_node=4 --master-port=$PORT main.py \
            --cfg "$CFG" --seed $SEED --tag "$TAG" > "logs/${DATASET}/${MODEL_NAME}.logs" 2>&1 &
        
        # Capture PID
        PID=$!
        PIDS+=($PID)
        echo "  Started $MODEL_NAME on $DATASET with PID: $PID"
    done
    
    # Wait for all models for this dataset to finish
    if [ ${#PIDS[@]} -gt 0 ]; then
        echo "  Waiting for all models on $DATASET to complete..."
        for PID in "${PIDS[@]}"; do
            wait $PID
        done
    fi
    
    echo "Completed all models for $DATASET"
    echo "--------------------------------"
done

echo "All experiments completed!"