def run_entropy(dataset_name,
save_dir,
k_list,
models,
feature_types,
second=False):
train_tokens, dev_tokens, train_dev_tokens, test_tokens, \
train_labels, dev_labels, train_dev_labels, test_labels = utils.load_data(dataset_name)
y_data, min_vals, max_vals, y_min_val, y_max_val = [], [], [], 0, 0
for model_name in models:
dicts, d_keys = als.create_model_d(save_dir,
model_name,
test_labels=test_labels)
tmp_y_data = als.get_entropy(test_tokens, dicts, d_keys, k_list)
assert len(tmp_y_data) == len(d_keys)
tmp_min_val, tmp_max_val = als.get_min_max(tmp_y_data)
min_vals.append(tmp_min_val)
max_vals.append(tmp_max_val)
y_data.append(tmp_y_data)
y_min_val = np.min(min_vals) - 0.25
y_min_val = max(0, y_min_val)
y_max_val = np.max(max_vals) + 0.25
simi.show_simi_plot(k_list, y_data, 'Number of important features (k)', 'Entropy', '', \
(13, 12), '', x_min=np.min(k_list)-0.5, x_max=np.max(k_list)+0.5, \
y_min=y_min_val, y_max=y_max_val, if_model=True, second=second, \
if_combi=False, if_builtin_posthoc=True)
def run_simi_length(dataset_name,
save_dir,
k_list,
models,
feature_types,
folder_name,
var,
second=False):
train_tokens, dev_tokens, train_dev_tokens, test_tokens, \
train_labels, dev_labels, train_dev_labels, test_labels = utils.load_data(dataset_name)
if var == 'len':
variable_l = als.get_tokens_length(test_tokens)
else:
variable_l = als.get_tokens_ratio(test_tokens)
# generate models line plot
y_data, min_vals, max_vals, y_min_val, y_max_val = [], [], [], 0, 0
for model_name in models:
combinations = als.get_model_combinations()
dicts, dict_keys = als.create_model_d(save_dir, model_name,
test_labels)
all_combi_data = als.get_rho(test_tokens, dicts, combinations, k_list,
variable_l)
tmp_min_val, tmp_max_val = als.get_min_max(all_combi_data)
min_vals.append(tmp_min_val)
max_vals.append(tmp_max_val)
y_data.append(all_combi_data)
y_min_val = np.min(min_vals) - 0.05
y_min_val = min(y_min_val, 0 - 0.05)
y_max_val = np.max(max_vals) + 0.05
y_max_val = max(y_max_val, 0 + 0.05)
simi.show_simi_plot(k_list, y_data, 'Number of important features (k)', 'Spearman correlation', '', \
(13, 12), '', x_min=np.min(k_list)-0.5, x_max=np.max(k_list)+0.5, \
y_min=y_min_val, y_max=y_max_val, if_model=True, second=second, \
if_builtin_posthoc=True)
def run_js_pos(dataset_name,
data_dir,
save_dir,
models,
feature_types,
k_list,
second=False):
train_tokens, dev_tokens, train_dev_tokens, test_tokens, \
train_labels, dev_labels, train_dev_labels, test_labels = utils.load_data(dataset_name)
train_pos, dev_pos, train_dev_pos, test_pos = utils.get_pos(
dataset_name, data_dir)
token_pos_d = als.get_token_pos_d(test_tokens, test_pos)
# compare with background
y_data, min_vals, max_vals, y_min_val, y_max_val = [], [], [], 0, 0
for model_name in models:
dicts, d_keys = als.create_model_d(save_dir,
model_name,
test_labels=test_labels)
tmp_y_data = get_jensen_shannon(test_tokens, dicts, d_keys, k_list, 'background', \
combinations=d_keys, token_pos_d=token_pos_d)
tmp_min_val, tmp_max_val = als.get_min_max(tmp_y_data)
min_vals.append(tmp_min_val)
max_vals.append(tmp_max_val)
y_data.append(tmp_y_data)
y_min_val = np.min(min_vals) - 0.05
y_max_val = np.max(max_vals) + 0.05
simi.show_simi_plot(k_list, y_data, 'Number of important features (k)', 'Jensen-Shannon Score', '', \
(13, 12), '', y_min=y_min_val, y_max=y_max_val, if_model=True, second=second, \
if_combi=False, if_background=True, if_builtin_posthoc=True)