A framework for modeling nucleic acid nearest neighbor motif energies with a CNN
usage: train_model.py [-h] [-o OPTIMIZER] [-e EPOCHS] [-b BATCH_SIZE]
[-p KEEPPROB] [-z REGULARIZE] [-l LEARNING_RATE]
[-m] [--batch_norm] [-r RESTORE] [-s] [--sterr]
[-g NUM_GPUS] [-t TESTFILE] [-c SCALE] [-n NORM]
[-d SEED] [-a ALPHA] [--linear] [--dilate]
filename n_units
train a CNN for nearest neighbor parameters
positional arguments:
filename name of parameter file
n_units numbers of units in hidden layers
optional arguments:
-h, --help show this help message and exit
-o OPTIMIZER, --optimizer OPTIMIZER
optimizer type (either "descent", "adam", "adagrad",
or "rmsprop")
-e EPOCHS, --epochs EPOCHS
number of epochs to train
-b BATCH_SIZE, --batch_size BATCH_SIZE
number of data points in each training batch
-p KEEPPROB, --keepprob KEEPPROB
probability to keep a node in dropout
-z REGULARIZE, --regularize REGULARIZE
regularization parameter for motifs
-l LEARNING_RATE, --learning_rate LEARNING_RATE
learning rate for model training
-m, --low_mem low memory mode
--batch_norm enable batch normalization
-r RESTORE, --restore RESTORE
file to restore weights from
-s, --save save log files and models
--sterr whether not to use standard error in loss function
-g NUM_GPUS, --num_gpus NUM_GPUS
number of gpus
-t TESTFILE, --testfile TESTFILE
test dataset
-c SCALE, --scale SCALE
scale factor between dG and dH
-n NORM, --norm NORM order of norm for loss function
-d SEED, --seed SEED random seed
-a ALPHA, --alpha ALPHA
regularization parameter for weights
--linear linear least squares fit
--dilate use dilated convlutions