def compare_tree_options(seed,
data,
tree_types,
parent_dir,
options,
data_amount,
prepend_attempt=""):
training_options_map = {
'training-options': util.myget(options, 'training-options')
}
combined_stats = {}
for t in tree_types:
options['tree-options'] = util.myget(all_tree_options, t)(util.myget(
options, 'tree-options', {}))
no_opts_stats = classifier_and_record(seed, data, t, parent_dir,
data_amount,
training_options_map,
prepend_attempt)
with_opts_stats = classifier_and_record(seed, data, t, parent_dir,
data_amount, options,
prepend_attempt)
stats = {}
stats['no-options'] = no_opts_stats
stats['with-options'] = with_opts_stats
# stats['with-options']['options'] = util.myget(options, 'tree-options', {})
combined_stats[t] = stats
return combined_stats
def wine_net(seed, r, nn_type, options=None):
if nn_type is 'cat':
filename = '../resources/winequality/winequality-combined-cat.csv'
y_key = 'Reviews'
split_options = {'exclude-keys': ['quality']}
else:
filename = '../resources/winequality/winequality-combined.csv'
y_key = 'quality'
split_options = {'exclude-keys': []}
split_options['perc_sample'] = r
dir = 'nn-exports/wine/'
seed_dir = 'attempts/' + dir + str(seed) + '/'
np.random.seed(seed)
options = util.default(options, {})
class_types = util.myget(options, 'class_types', ['net'])
layers = util.myget(options, 'layers', [(5, 6)])
data = util.get_data(filename)
split_data = util.prep_data(data, y_key, split_options)
net_results = {}
for t in class_types:
for l in layers:
np.random.seed(seed)
start_time = datetime.datetime.now()
[train_results,
test_results] = train_and_test(seed_dir, split_data, t, l)
net_results[t + '_train'] = train_results
net_results[t + '_test'] = test_results
total_time = datetime.datetime.now() - start_time
write_results(seed, t, dir, train_results['error-perc'],
test_results['error-perc'], total_time,
len(split_data[0]), l, nn_type + '-')
pass
def breast_cancer_net(seed, r, options=None):
dir = 'nn-exports/breast_cancer/'
seed_dir = 'attempts/' + dir + str(seed) + '/'
np.random.seed(seed)
options = util.default(options, {})
class_types = util.myget(options, 'class_types', ['net'])
layers = util.myget(options, 'layers', [(5, 6)])
data = util.get_data('../resources/breast-cancer/wdbc.data.csv')
data = util.split_feature(data, 'diagnosis')
split_data = util.prep_data(
data, 'diagnosis_M', {
'exclude-keys': ['id', 'diagnosis_B', 'diagnosis'],
'perc_sample': r
})
net_results = {}
for t in class_types:
for l in layers:
np.random.seed(seed)
start_time = datetime.datetime.now()
[train_results,
test_results] = train_and_test(seed_dir, split_data, t, l)
net_results[t + '_train'] = train_results
net_results[t + '_test'] = test_results
total_time = datetime.datetime.now() - start_time
write_results(seed, t, dir, train_results['error-perc'],
test_results['error-perc'], total_time,
len(split_data[0]), l)
return net_results
def classify(seed, data, tree_type, amount_data, options=None):
np.random.seed(seed)
tree_options = util.myget(options, 'tree-options', {})
x_train, x_test, y_train, y_test, n_test = prep_data(
data, util.myget(options['training-options'], 'y-key'),
options['training-options'], amount_data)
classifier = tree_type(**tree_options)
classifier.fit(x_train, y_train)
return classifier, x_train, x_test, y_train, y_test, n_test
def decision_classifier_options(options=None):
max_leaf_nodes = util.myget(options, 'max-leaf-nodes')
max_depth = util.myget(options, 'max-depth')
return {
'class_weight': None,
'criterion': 'gini',
'max_depth': max_depth,
'max_leaf_nodes': max_leaf_nodes,
'min_samples_leaf': 5,
'min_samples_split': 2,
'min_weight_fraction_leaf': 0.0,
'presort': False,
'random_state': None,
'splitter': 'random'
}
def write_results(seed,
tree_type,
data_dir,
data_amount,
n_test,
error_train,
error_test,
time,
options=None,
prepend_attempt=""):
sub = '-with-opts' if util.myget(options, 'tree-options',
None) is not None else '-no-opts'
fields = [
attempt, tree_type + sub, seed, data_amount, n_test, error_train,
error_test, time,
str(options)
]
with open(data_dir + prepend_attempt + attempt + '-results-comb.csv',
'a') as f:
writer = csv.writer(f)
writer.writerow(fields)
with open(
data_dir + prepend_attempt + attempt + '-results-' + tree_type +
sub + '.csv', 'a') as f:
writer = csv.writer(f)
writer.writerow(fields)
def prep_data(data, y_key, options=None, amount_data=0.67):
options = util.default(options, {})
exclude_keys = util.myget(options, 'exclude-keys', [])
exclude_keys.append(y_key)
x = data.loc[:, data.columns.difference(exclude_keys)]
y = data.loc[:, y_key]
x_train, x_test, y_train, y_test = train_test_split(x,
y,
test_size=1 -
amount_data)
return x_train, x_test, y_train, y_test, len(x_train)
def classifier_and_record(seed,
data,
t,
parent_dir,
data_amount,
options=None,
prepend_attempt=""):
start_time = datetime.datetime.now()
tree_type = util.myget(all_tree_types, t)
# Creates and trains a classifier, then saves charts and stats
tree_options = util.myget(options, 'tree-options')
fn = 'basic' if tree_options is None else 'option'
tree_filename = util.myget(util.myget(options, 'training-options'),
'image-name').format('tree', t, fn, 'svg')
classifier, x_train, x_test, y_train, y_test, n_test = classify(
seed, data, tree_type, data_amount, options)
if t is not 'ada':
charts.save_tree_chart(tree_filename, classifier)
results = [
performance(
classifier, x_train, y_train,
util.myget(util.myget(options, 'training-options'),
'image-name').format('confusion-matrix-train', t, fn,
'csv')),
performance(
classifier, x_test, y_test,
util.myget(util.myget(options, 'training-options'),
'image-name').format('confusion-matrix-test', t, fn,
'csv'))
]
[train_results, test_restuls] = results
total_time = datetime.datetime.now() - start_time
write_results(seed, t, parent_dir, data_amount, n_test,
train_results['error-perc'], test_restuls['error-perc'],
total_time, options, prepend_attempt)
return results