A platform for training and testing convolutional neural network (CNN) based on Caffe

Caffe and pycaffe (already included for mri-wrapper)

First replace all $REPO_HOME$ in the following files as the actual directory of the repository:

• example/param.list
• example/data/train.txt
• example/data/valid.txt
• example/data/test.txt

python run.py train example/param.list

python run.py test example/param.list

python run.py test_eval example/param.list

The output will be under /$REPO_HOME$/example/basicmodel/version0/best_trial/

All the data should be put under $model_topdir$/data/, where model_topdir is the top folder of the output.

An example list of data is here

For mri-wrapper user, we now provide a python script to automatically generate the corresponding data matrix and data manifest file from 1-D genomic sequence in FASTA or TSV format. The user still has to manually split the sequence data into train, valid and test set before using this script on each one of them. Check out the usage of the script by:

python embedH5.py -h

Training, validating and testing data should be each prepared as 4-D matrix with dimension number * channel * width * height in HDF5 format. A good practice is to partition the dataset into small batches.

For each of train/valid/test set, a train.txt/valid.txt/test.txt file is required to specify the ABSOLUTE PATH of all the HDF5 batches in the set.

Refer to here and here for instructions and examples in generating the following three files:

• trainval.prototxt: architecture for training
• solver.prototxt: solver parameters
• deploy.prototxt: architecture for testing

Note in trainval.prototxt

• Data must be input through a hdf5 layer
• Data source (for training as an example) should be either "../../../data/train.txt" or an absolute path $model_topdir$/data/train.txt.

This file specificies all the parameters in the model.

Example : (example/param.list)

model_topdir /$REPO_HOME$/example/
data_src NA
gpunum 0
solver_file example/solver.prototxt
trainval_file example/trainval.prototxt
deploy_file example/deploy.prototxt
codedir $REPO_HOME$
modelname basicmodel
model_batchname version0
predictmodel_batch trail0
predict_filelist data/test.txt
trial_num 3
optimwrt accuracy

• model_topdir: The top output folder
• data_src: not used in current version
• gpunum: The GPU device number to run on
• solver_file: The path to solver file
• trainval_file: The path to trainval file
• deploy_file: The path to deploy file
• codedir: The directory of this repo
• modelname: The name of the model.
• model_batchname: The name of the specific model version. All the output will be therefore generated under $model_topdir$/$modelname$/$model_batchname$/, i.e. /cluster/output/basicmodel/trail0/ in the example.
• predictmodel_batch: The model version to test on
• predict_filelist: The path of the test HDF5 file relative to model_topdir
• trial_num: The number of training trial.
• optimwrt: choose the best trial wrt this metric (accuracy or loss)

Train a CNN using the parameters provided in param.list. $trial_num$ number of independent training will be performed using the same parameters. The output will be under $model_topdir$/$modelname$/$model_batchname$/

python run.py train param.list

First, from all the model files saved across different trials and iterations, the one with best validation accuracy will be picked. Then input the test data into this model to generate prediction output. All the ouput in test phase are under /$model_topdir$/$modelname$/$model_batchname$/best_trial/

python run.py test param.list

Evaluate the performance on test set by accuracy and area under receiver operating curve (auROC). The output is in the same output folder as test phase.

python run.py test_eval param.list