def structured_output_quality(links) -> (List, float, float, float):
"""
Infer component labels automatically from the structure
"""
component_labels = []
tree_ratio = 0
avg_depth = 0
avg_leaf_prop = 0
all_depths = []
n_essays = len(links)
for i in range(len(links)):
rep = TreeBuilder(links[i])
component_labels.append(rep.auto_component_labels(AC_breakdown=True))
if rep.is_tree():
tree_ratio += 1
# evaluate this only when the output forms a tree
depth, leaf_prop = rep.tree_depth_and_leaf_proportion()
avg_depth += depth
all_depths.append(depth)
avg_leaf_prop += leaf_prop
return component_labels, float(tree_ratio) / float(n_essays), float(
avg_depth) / float(tree_ratio), float(avg_leaf_prop) / float(
tree_ratio), all_depths
# to get non_AC samples
non_ac_samples.extend(essay.get_non_ACS("original", False))
# about tree structure
if args.original_order:
directions_with_non_AC = essay.get_rel_distances(
"original", include_non_arg_units=True)[0]
else:
directions_with_non_AC = essay.get_rel_distances(
"reordering", include_non_arg_units=True)[0] # reordering
try:
rep = TreeBuilder(
directions_with_non_AC) # distances between sentences
except:
print("Distance error", essay.essay_code)
depth, leaf_ratio = rep.tree_depth_and_leaf_proportion()
leaf_prop.append(leaf_ratio)
print("> Corpus", directory)
print("> items", len(essays))
print("> Common Stats")
print(" \t\t\t \tsum \tmax \tmin \tavg \tstdev")
print_stats("# Sentences\t", n_sentences)
print_stats("# Tokens\t", n_tokens)
print_stats("# Arg. components", n_ACs)
print_stats("# Non-arg. comp.", n_non_ACs)
print("> Relations")
print_stats("# Support\t", n_sup)
print_stats("# Detail\t", n_det)
print_stats("# Attack\t", n_att)
