Running trainmodel
on the command line¶
Basic instructions¶
Basic instructions on running trainmodel
can be gained by looking at the --help
menu:
$ trainmodel --help
Using Theano backend.
WARNING (theano.configdefaults): install mkl with `conda install mkl-service`: No module named 'mkl'
usage: trainmodel [-h] [--path-to-trainingset PATH_TO_TRAININGSET]
--number-of-classes NUMBER_OF_CLASSES
[--trainingset-pickle-file TRAININGSET_PICKLE_FILE]
[--model-name MODEL_NAME] [--batch-size BATCH_SIZE]
[--nb-epoch NB_EPOCH]
[--fraction-validation FRACTION_VALIDATION]
[--fraction-testing FRACTION_TESTING]
[--randomseed RANDOMSEED] [--verbose]
An examples commandline of how to obtain a model is given below:
THEANO_FLAGS=mode=FAST_RUN,device=cuda,floatX=float32 trainmodel --path-to-
trainingset='somedir' --number-of-classes='somenum'
optional arguments:
-h, --help show this help message and exit
--path-to-trainingset PATH_TO_TRAININGSET
folder where labeled images live
--number-of-classes NUMBER_OF_CLASSES
How many classes do you have
--trainingset-pickle-file TRAININGSET_PICKLE_FILE
folder where the entire pickled training set will
live. This pickle file should be read in by pandas
--model-name MODEL_NAME
what you would like to model filename to be
--batch-size BATCH_SIZE
defines the batch size, 30 is a reasonable size
--nb-epoch NB_EPOCH defines the number of iterations, 130 is reasonable.
You can set it to 100 or below, if you have time
concern for training.
--fraction-validation FRACTION_VALIDATION
Perentage of trianing set to save for validation
--fraction-testing FRACTION_TESTING
Percentage of training set to save for testing
--randomseed RANDOMSEED
Set random seed
--verbose Run in Verbose Mode
Detailed instructions¶
A few of the command-line options require special formatting, use the reference below for more detailed info.