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

'''
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 fixed for this experiment
codebase_clone_path = "/u/savardf/ift6266/experiment_clones/ift6266_experiment10"

# 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])

# 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_experiment10.ift6266.deep.stacked_dae.nist_sda.jobman_entrypoint"

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

from jobman import DD

'''
These are parameters used by nist_sda.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.
'''

# 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 = False

NIST_ALL_LOCATION = '/data/lisa/data/nist/by_class/all'
NIST_ALL_TRAIN_SIZE = 649081
# valid et test =82587 82587 

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

# reduce training set to that many examples
REDUCE_TRAIN_TO = None
# that's a max, it usually doesn't get to that point
MAX_FINETUNING_EPOCHS = 1000
# number of minibatches before taking means for valid error etc.
REDUCE_EVERY = 100

if TEST_CONFIG:
    REDUCE_TRAIN_TO = 1000
    MAX_FINETUNING_EPOCHS = 2
    REDUCE_EVERY = 10


# 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 = {'pretraining_lr': [0.1, 0.01],#, 0.001],#, 0.0001],
        'pretraining_epochs_per_layer': [10,20],
        'hidden_layers_sizes': [300,800],
        'corruption_levels': [0.1,0.2,0.3],
        'minibatch_size': [20],
        'max_finetuning_epochs':[MAX_FINETUNING_EPOCHS],
        'finetuning_lr':[0.1, 0.01], #0.001 was very bad, so we leave it out
        'num_hidden_layers':[2,3]}

# Just useful for tests... minimal number of epochs
# (This is used when running a single job, locally, when
# calling ./nist_sda.py test_jobman_entrypoint
DEFAULT_HP_NIST = DD({'finetuning_lr':0.1,
                       'pretraining_lr':0.1,
                       'pretraining_epochs_per_layer':2,
                       'max_finetuning_epochs':2,
                       'hidden_layers_sizes':800,
                       'corruption_levels':0.2,
                       'minibatch_size':20,
                       'reduce_train_to':10000,
                       'num_hidden_layers':1})

# To reinsert duplicate of jobs that crashed
REINSERT_COLS = ['pretraining_lr','pretraining_epochs_per_layer','hidden_layers_sizes','corruption_levels','minibatch_size','max_finetuning_epochs','finetuning_lr','num_hidden_layers','dupe']
REINSERT_JOB_VALS = [\
            [0.1,10,800,0.3,20,1000,0.01,3,2],
            [0.1,10,800,0.4,20,1000,0.01,3,2],
            [0.1,10,800,0.3,20,1000,0.005,3,2],
            [0.1,10,800,0.6,20,1000,0.005,3,2]]