def generate_dif_mat_single(predictions, labels, plot_flg=False, sub_set=""):
"""
Generates diffusion matrix for single output
:param predictions:
:param labels:
:param plot_flg:
:param sub_set:
:return:
"""
matrices = []
# single case
for i in range(1):
s = (np.shape(predictions)[-1], np.shape(predictions)[-1])
matrices.append(np.zeros(shape=s))
for pred, lab in zip(predictions, labels):
p = np.argmax(pred)
matrices[0][p][lab] += 1
if plot_flg:
for i in range(1):
plot_matrix(matrices[i],
str(i) + "_" + sub_set + "_",
get_cat_attributes_names())
return matrices
def generate_dif_mat(predictions, labels, plot_flg=False, sub_set=""):
matrices = []
att_cnt = len(predictions)
#todo add iff predictions empty
for i in range(att_cnt):
l = len(predictions[i][0])
s = (l, l)
print("Shape:", s)
matrices.append(np.zeros(shape=s))
for att_pred, att_lab, i in zip(predictions, labels, range(att_cnt)):
for pred, lab in zip(att_pred, att_lab):
if lab != MASK_VALUE:
# categorical
p = np.argmax(pred)
# l = np.argmax(lab)
l = lab
# matrices[i][p][l] += 1
# sparse
matrices[i][p][l] += 1
if plot_flg:
for i in range(att_cnt):
plot_matrix(matrices[i],
str(i) + "_" + sub_set + "_",
get_cat_attributes_names())
return matrices
def plot_diff_matrices(matrices, split_name, ind=None):
"""
Plots percentage diffusion matrices and saves them as .png files
into figures/confusion folder
:param matrices:
:param split_name: train/test/validation, it will be used as prefix
for saved images
:return:
"""
if ind is None:
names = get_cat_attributes_names()
categories = get_category_names()
else:
names = [get_cat_attributes_names()[ind]]
categories = [get_category_names()[ind]]
for matrix, alpha, cat in zip(matrices, names, categories):
print("-----------------", split_name)
plot_matrix(convert_to_percentage_mat(matrix), split_name + "_" + cat,
alpha)
def run_err_stats():
"""
Computes error rate in percentage per category.
Requires diff_dict.npy to load data.
:return:
"""
d_d = load_dictionary("diff_dict.npy")
alphas = get_cat_attributes_names()
for key in d_d.keys():
print("----- ", key, "-------")
for matrix, alpha in zip(d_d[key], alphas):
print_errs(convert_to_percentage_mat(matrix), alpha)
def plot_diff_matrices_m(matrices, split_name):
"""
Plots percentage diffusion matrices and saves them as .png files
into figures/confusion folder
:param matrices:
:param split_name: train/test/validation, it will be used as prefix
for saved images
:return:
"""
names = get_cat_attributes_names()
categories = get_category_names()
names.append(["Adolescent", "Adult", "Middle aged", "Retired", "Old"])
categories.append("Age")
for matrix, alpha, cat in zip(matrices, names, categories):
plot_matrix(convert_to_percentage_mat(matrix), split_name + "_" + cat,
alpha)