coder02lq/geformer repository overview

⁠GEFormer

⁠1. Quick Start

⁠1.1 Pull Docker Image

docker pull coder02lq/geformer:v1.1

⁠Data Introduction

⁠CVF File Examples

Cross-validation files for splitting training and validation sets.

First column: ID, Second column: CV

M1_CVF.csv 【Ten fold cross validation】

M2_CVF.csv 【K-fold cross validation (K=number of environments)】

M3_CVF.csv 【Ten*K fold cross validation (K=number of environments)】

⁠Environment File Examples

Environmental data with fixed first two column names (env and date), followed by environmental factor data from the third column.

File Name：${envName}_env.csv

LN_env.csv

⁠Genotype File Examples

Genotypic data with first column name: ID

geno.csv

⁠Phenotype File Examples

Phenotypic data

File Name：${phenoName}_phe.csv

Column Name：${phenoName}_${envName}

PH_phe.csv

⁠Parameter Introduction

⁠Required Parameters

• --geno_path // Path to genotype data file
• --phe_folder // Folder containing phenotype data
• --pheno_name // Phenotype name
• --env_folder // Folder containing environment data
• --env_name // Environment name required for M1 and M3 schemes
• --cvf_folder // Folder for training/validation set configuration
• --scheme // Choose M1/M2/M3 scheme
• --device // Which GPU to use
• --need_optuna // Whether to enable hyperparameter optimization

⁠Hyperparameters

When --need_optuna is True: Automatically optimize combinations of the following hyperparameters within specified ranges

When --need_optuna is False: Use default parameters or specify the following parameters

⁠Basic Parameters

• --batch // Batch size (default: 64)
• --dropout // Dropout rate (prevents overfitting) (default: 0.3)
• --depth // Number of feature layers (default: 2)
• --neurons1 // Number of neurons 1 (default: 256)
• --neurons2 // Number of neurons 2 (default: 32)
• --lr // Learning rate (default: 5e-4)

⁠Optuna Optimization Parameters

• --need_optuna // Whether to enable hyperparameter optimization (default: True)
• --optuna_epoch // Number of attempts with different parameter combinations (default: 100)
• --num_fold // Folded number (default: 1)

⁠Optuna Search Ranges

• --batch_1 // Minimum batch size (default: 16)
• --batch_2 // Maximum batch size (default: 128)
• --dropout_1 // Minimum dropout rate (default: 0.2)
• --dropout_2 // Maximum dropout rate (default: 0.6)
• --depth_1 // Minimum depth (default: 1)
• --depth_2 // Maximum depth (default: 6)
• --neurons1_1 // Minimum neurons1 number (default: 128)
• --neurons1_2 // Maximum neurons1 number (default: 512)
• --neurons2_1 // Minimum neurons2 number (default: 1)
• --neurons2_2 // Maximum neurons2 number (default: 128)
• --lr_1 // Minimum learning rate (default: 1e-7)
• --lr_2 // Maximum learning rate (default: 1e-2)

⁠Usage Examples

⁠1. Manual Path Specification (Advanced)

If your data structure differs from default, map volumes manually:

# Example for Manual Mapping
docker run --gpus all --rm -it \
  -v "/your/local/data:/data_input" \
  -v "/your/local/best_model:/gef/code/best_model" \
  -v "/your/local/k_model:/gef/code/k_model" \
  -w /gef/code \
  coder02lq/geformer:v1.1 \
  python train.py \
  --scheme M1 \
  --geno_path /data_input/geno/geno.csv \
  --phe_folder /data_input/phe/ \
  --env_folder /data_input/env \
  --cvf_folder /data_input/cvf \
  --env_name LN \
  --pheno_name PH

⁠2. Simplified (Recommended)

If your data folder matches standard structure (contains geno, phe, env folders):

docker run --gpus all --rm -it \
  -v "/your/local/data:/gef/data" \
  -v "/your/local/best_model:/gef/code/best_model" \
  -v "/your/local/k_model:/gef/code/k_model" \
  -w /gef/code \
  coder02lq/geformer:v1.1 \
  python train.py --scheme M1 --env_name LN --pheno_name PH

⁠3. Command Parameters Explanation

Note: The simplified command works because the code defaults to reading from ../data (/gef/data in container).