def __init__(self,
file_prefix,
users_as_tasks=False,
num_cross_folds=DEFAULT_NUM_CROSS_FOLDS,
cont=False,
results_path=DEFAULT_RESULTS_PATH,
figures_path=DEFAULT_FIGURES_PATH,
datasets_path=DEFAULT_DATASETS_PATH,
test_run=False,
max_iters=DEFAULT_MAX_ITERS,
val_type=DEFAULT_VALIDATION_TYPE,
optimize_labels=None,
test_csv_filename=None):
self.results_path = results_path
self.figures_path = figures_path
self.datasets_path = datasets_path
self.save_prefix = self.getSavePrefix(file_prefix, replace=cont)
self.cont = cont
self.max_iters = max_iters
self.val_type = val_type
self.users_as_tasks = users_as_tasks
self.file_prefix = file_prefix
if test_csv_filename is not None:
self.test_csv_filename = self.datasets_path + test_csv_filename
else:
self.test_csv_filename = None
self.test_tasks = helper.loadPickledTaskList(datasets_path,
file_prefix, "Test")
self.train_tasks = helper.loadPickledTaskList(datasets_path,
file_prefix, "Train")
if self.val_type != 'cross':
self.val_tasks = helper.loadPickledTaskList(
datasets_path, file_prefix, "Val")
self.initializeHBLRModel(self.train_tasks)
else:
self.classifier = None
if users_as_tasks:
self.K = 25
else:
self.K = len(self.test_tasks)
self.n_feats = helper.calculateNumFeatsInTaskList(self.test_tasks)
self.n_tasks = len(self.test_tasks)
if optimize_labels is None:
self.optimize_labels = [
'tomorrow_Group_Happiness_Evening_Label',
'tomorrow_Group_Health_Evening_Label',
'tomorrow_Group_Calmness_Evening_Label'
]
else:
self.optimize_labels = optimize_labels
#parameters that can be tuned
self.tau10s = [10, 1, 0.05, 0.01]
self.tau20s = [1.0, 0.05, 0.01]
self.sigma_multipliers = [.01, 0.1, 1]
self.mu_multipliers = [0.0]
if test_run:
print "This is only a testing run. Using cheap settings to make it faster"
self.K = 2
self.max_iters = 5
self.n_tasks = 2
self.tau10s = [1]
self.tau20s = [.1]
self.sigma_multipliers = [.01]
self.mu_multipliers = [0]
self.calcNumSettingsDesired()
#storing the results
self.time_sum = 0
if cont:
self.val_results_df = pd.DataFrame.from_csv(self.results_path +
self.save_prefix +
'.csv')
print '\nPrevious validation results df loaded. It has', len(
self.val_results_df), "rows"
self.started_from = len(self.val_results_df)
else:
self.val_results_df = pd.DataFrame()
self.started_from = 0
self.num_cross_folds = num_cross_folds
if self.val_type == 'cross':
helper.generateCrossValPickleFiles(self.datasets_path,
self.file_prefix,
self.num_cross_folds)
def __init__(self,
file_prefix,
users_as_tasks,
user_clusters=True,
eta_filename=None,
regularizers=REGULARIZERS,
tolerance=.0001,
max_iter=100,
val_type=VALIDATION_TYPE,
c_vals=C_VALS,
beta_vals=B_VALS,
v_vals=V_VALS,
kernels=KERNELS,
print_iters=False,
optimize_labels=None,
cont=False,
test_run=False,
results_path=DEFAULT_RESULTS_PATH,
figures_path=DEFAULT_FIGURES_PATH,
datasets_path=DEFAULT_DATASETS_PATH,
etas_path=DEFAULT_ETAS_PATH,
num_cross_folds=DEFAULT_NUM_CROSS_FOLDS,
drop20=False,
test_csv_filename=None):
self.results_path = results_path
self.figures_path = figures_path
self.datasets_path = datasets_path
self.etas_path = etas_path
self.file_prefix = file_prefix
self.cont = cont
self.val_type = val_type
self.users_as_tasks = users_as_tasks
self.cluster_users = user_clusters
self.drop20 = drop20
if test_csv_filename is not None:
self.test_csv_filename = self.datasets_path + test_csv_filename
else:
self.test_csv_filename = None
self.save_prefix = self.getSavePrefix(file_prefix, replace=cont)
self.test_tasks = helper.loadPickledTaskList(datasets_path,
file_prefix,
"Test",
fix_y=True)
self.train_tasks = helper.loadPickledTaskList(datasets_path,
file_prefix,
"Train",
fix_y=True)
if self.val_type != 'cross':
self.val_tasks = helper.loadPickledTaskList(datasets_path,
file_prefix,
"Val",
fix_y=True)
# print dataset sizes
print "Num train points:", sum([len(t['Y']) for t in self.train_tasks])
if self.val_type != 'cross':
print "Num val points:", sum([len(t['Y']) for t in self.val_tasks])
print "Num test points:", sum([len(t['Y']) for t in self.test_tasks])
if self.val_type != 'cross':
self.initializeMTMKLModel(self.train_tasks)
else:
self.classifier = None
self.n_feats = helper.calculateNumFeatsInTaskList(self.test_tasks)
self.n_tasks = len(self.test_tasks)
if optimize_labels is None:
self.optimize_labels = [
'tomorrow_Group_Happiness_Evening_Label',
'tomorrow_Group_Health_Evening_Label',
'tomorrow_Group_Calmness_Evening_Label'
]
else:
self.optimize_labels = optimize_labels
self.c_vals = c_vals
self.v_vals = v_vals
self.kernels = kernels
self.beta_vals = beta_vals
self.regularizers = regularizers
self.tolerance = tolerance
self.max_iter = max_iter
self.print_iters = print_iters
if test_run:
print "This is only a testing run. Using cheap settings to make it faster"
self.c_vals = [100]
self.beta_vals = [.01]
self.kernels = ['linear']
self.v_vals = [1.0]
self.regularizers = ['L1']
self.max_iter = 1
self.calcNumSettingsDesired()
#storing the results
self.time_sum = 0
if cont:
self.val_results_df = pd.DataFrame.from_csv(self.results_path +
self.save_prefix +
'.csv')
print '\nPrevious validation results df loaded. It has', len(
self.val_results_df), "rows"
self.started_from = len(self.val_results_df)
else:
self.val_results_df = pd.DataFrame()
self.started_from = 0
self.num_cross_folds = num_cross_folds
if self.val_type == 'cross':
helper.generateCrossValPickleFiles(self.datasets_path,
self.file_prefix,
self.num_cross_folds)
def __init__(self,
file_prefix,
users_as_tasks=False,
cont=False,
classifier_name='LSSVM',
num_cross_folds=DEFAULT_NUM_CROSS_FOLDS,
main_directory=DEFAULT_MAIN_DIRECTORY,
datasets_path='Data/Datasets/Discard20/',
cant_train_with_one_class=True,
check_test=False,
save_results_every_nth=3,
test_csv_filename=None):
""" Initializes the parent model with fields useful for all child wrapper classes
Args:
file_prefix: The first portion of the name of a set of pickled task lists, e.g.
'datasetTaskList-Discard-Future-Group_'
users_as_tasks: A boolean. If true, will assume there are many tasks and each task
is one person. Will not print results per task.
cont: A boolean. If true, will try to load a saved results .csv and continue
training on the next unfinished result.
classifier_name: String name of the classifier trained. Used to know where to save
results.
num_cross_folds: An integer number of folds to use in cross validation.
main_directory: The path to the main dropbox directory which contains the results and
data directories.
datasets_path: The path from the main dropbox to the datasets directory.
cant_train_with_one_class: A boolean. If true, if the model encounters a task with
only one type of label in the training data, it will just predict the most
frequent class.
check_test: A boolean. If true, will evaluate final results on held-out test set
after running.
save_results_every_nth: An integer representing the number of settings to test before
writing the results df to a csv file.
"""
# memorize arguments and construct paths
self.main_directory = main_directory
self.classifier_name = classifier_name
self.results_path = main_directory + 'Results/' + classifier_name + '/'
self.figures_path = main_directory + 'Figures/' + classifier_name + '/'
self.datasets_path = main_directory + datasets_path
self.cont = cont
self.users_as_tasks = users_as_tasks
self.cant_train_with_one_class = cant_train_with_one_class
self.check_test = check_test
self.save_results_every_nth = save_results_every_nth
self.file_prefix = file_prefix
self.save_prefix = self.get_save_prefix(file_prefix, replace=cont)
if test_csv_filename is not None:
self.test_csv_filename = self.datasets_path + test_csv_filename
else:
self.test_csv_filename = None
self.params = {}
self.define_params()
self.load_data()
self.calc_num_param_settings()
self.construct_list_of_params_to_test()
#storing the results
self.time_sum = 0
if cont:
self.val_results_df = pd.DataFrame.from_csv(self.results_path +
self.save_prefix +
'.csv')
print '\nPrevious validation results df loaded. It has', len(
self.val_results_df), "rows"
self.started_from = len(self.val_results_df)
else:
self.val_results_df = pd.DataFrame()
self.started_from = 0
self.num_cross_folds = num_cross_folds
helper.generateCrossValPickleFiles(self.datasets_path,
self.file_prefix,
self.num_cross_folds)