view deep/crbm/mnist_config.py.example @ 643:24d9819a810f

reviews aistats finales
author Yoshua Bengio <bengioy@iro.umontreal.ca>
date Thu, 24 Mar 2011 17:04:38 -0400
parents 1e9788ce1680
children
line wrap: on
line source

# ----------------------------------------------------------------------------
# BEGIN EXPERIMENT ISOLATION CODE

# Path to pass to jobman sqlschedule. IMPORTANT TO CHANGE TO REFLECT YOUR CLONE.
# Make sure this is accessible from the default $PYTHONPATH (in your .bashrc)
# (and make sure every subdirectory has its __init__.py file)
EXPERIMENT_PATH = "ift6266_mnistcrbm_exp1.ift6266.deep.crbm.mnist_crbm.jobman_entrypoint"

def isolate_experiment():
    '''
    This makes sure we use the codebase clone created for this experiment.
    I.e. if you want to make modifications to the codebase but don't want your
    running experiment code to be impacted by those changes, first copy the
    codebase somewhere, and configure this section. It will make sure we import
    from the right place.

    MUST BE DONE BEFORE IMPORTING ANYTHING ELSE
    (Leave this comment there so others will understand what's going on)
    '''

    # Place where you copied modules that should be frozen for this experiment
    codebase_clone_path = "/u/savardf/ift6266/experiment_clones/ift6266_mnistcrbm_exp1"

    # Places where there might be conflicting modules from your $PYTHONPATH
    remove_these_from_pythonpath = ["/u/savardf/ift6266/dev_code"]

    import sys
    sys.path[0:0] = [codebase_clone_path]

    # remove paths we specifically don't want in $PYTHONPATH
    for bad_path in remove_these_from_pythonpath:
        sys.path[:] = [el for el in sys.path if not el in (bad_path, bad_path+"/")]

    # Make the imports
    import ift6266

    # Just making sure we're importing from the right place
    modules_to_check = [ift6266]
    for module in modules_to_check:
        if not codebase_clone_path in module.__path__[0]:
            raise RuntimeError("Module loaded from incorrect path "+module.__path__[0])

# END EXPERIMENT ISOLATION CODE
# ----------------------------------------------------------------------------

from jobman import DD

'''
These are parameters used by mnist_crbm.py. They'll end up as globals in there.

Rename this file to config.py and configure as needed.
DON'T add the renamed file to the repository, as others might use it
without realizing it, with dire consequences.
'''

# change "sandbox" when you're ready
JOBDB = 'postgres://ift6266h10@gershwin/ift6266h10_sandbox_db/yourtablenamehere'

# Set this to True when you want to run cluster tests, ie. you want
# to run on the cluster, many jobs, but want to reduce the training
# set size and the number of epochs, so you know everything runs
# fine on the cluster.
# Set this PRIOR to inserting your test jobs in the DB.
TEST_CONFIG = False

# save params at training end
SAVE_PARAMS = True

IMAGE_OUTPUT_DIR = 'img/'

# number of minibatches before taking means for valid error etc.
REDUCE_EVERY = 100

# print series to stdout too (otherwise just produce the HDF5 file)
SERIES_STDOUT_TOO = False

# every X minibatches
VISUALIZE_EVERY = 1000 # x20, ie. every 20,000 examples
GIBBS_STEPS_IN_VIZ_CHAIN = 1000

if TEST_CONFIG:
    REDUCE_EVERY = 10
    VISUALIZE_EVERY = 20

# This is to configure insertion of jobs on the cluster.
# Possible values the hyperparameters can take. These are then
# combined with produit_cartesien_jobs so we get a list of all
# possible combinations, each one resulting in a job inserted
# in the jobman DB.
JOB_VALS = {'learning_rate': [1.0, 0.1, 0.01],
        'sparsity_lambda': [3.0,0.5],
        'sparsity_p': [0.3,0.05],
        'num_filters': [40,15],
        'filter_size': [12,7],
        'minibatch_size': [20],
        'num_epochs': [20]}

# Just useful for tests... minimal number of epochs
# Useful when launching a single local job
DEFAULT_STATE = DD({'learning_rate': 0.1,
        'sparsity_lambda': 1.0,
        'sparsity_p': 0.05,
        'num_filters': 40,
        'filter_size': 12,
        'minibatch_size': 10,
        'num_epochs': 20})

# To reinsert duplicate of jobs that crashed
REINSERT_COLS = ['learning_rate','sparsity_lambda','sparsity_p','num_filters','filter_size','minibatch_size','dupe']
#REINSERT_JOB_VALS = [\
#            [,2],]