# GENERATE INITIAL POPULATION
python3 popu_generator.py
# GENERATE ARCHIVE
# SINGLE GPU TRAINING
CUDA_VISIBLE_DEVICES=0 python3 detectorcube/code/archive.py
# MULTI GPU TRAINING
CUDA_VISIBLE_DEVICES=0,1,2,3 python3 detectorcube/code/archive.py
# FIT SURROGATES
# SINGLE GPU TRAINING
CUDA_VISIBLE_DEVICES=0 python3 surrrogates_fit.py
# MULTI GPU TRAINING
CUDA_VISIBLE_DEVICES=0,1,2,3 python3 surrogates_fit.py
# RUN SEARCh
# SINGLE GPU TRAINING
CUDA_VISIBLE_DEVICES=0 python3 nasod_gen_search.py
# MULTI GPU TRAINING
CUDA_VISIBLE_DEVICES=0,1,2,3 python3 nasod_gen_search.py
# TRAIN GENOMES
# SINGLE GPU TRAINING
CUDA_VISIBLE_DEVICES=0 python3 train_nasod.py
# MULTI GPU TRAINING
CUDA_VISIBLE_DEVICES=0,1,2,3 python3 train_nasod.py
# GET MEAN AVERAGE PRECISION
# SINGLE GPU TRAINING
CUDA_VISIBLE_DEVICES=0 python3 evaluate.py
# MULTI GPU TRAINING
CUDA_VISIBLE_DEVICES=0,1,2,3 python3 evaluate.py
# PLOT BBOX
# SINGLE GPU TRAINING
CUDA_VISIBLE_DEVICES=0 python3 plot_prediction.py
# MULTI GPU TRAINING
CUDA_VISIBLE_DEVICES=0,1,2,3 python3 plot_prediction.py
In an effort to efficiently map surrogate values to genomic data, I utilized an autoencoder to compress 18-bit genomic sequences into a single-bit representation. Initially, I trained the autoencoder on the task of reconstructing 18-bit genomic sequences, ensuring that the encoder and decoder learned to effectively compress and decompress the data. After achieving satisfactory reconstruction accuracy, I removed the decoder, retaining only the encoder. This encoder now serves as a tool to transform the 18-bit genomic sequences into a more manageable 1-bit format, facilitating the fitting of surrogates against the mapped values. This dimensionality reduction not only simplifies the data but also maintains the essential genetic information needed for accurate surrogate modeling.
