def create_ARSC_train_baseline_episode():
labels = sorted(
set([
line.strip() for line in open(
"data/ARSC-Yu/raw/workspace.filtered.list", "r").readlines()
]) - set([
line.strip() for line in open(
"data/ARSC-Yu/raw/workspace.target.list", "r").readlines()
]))
# Pick a random label
label = random.choice(labels)
# Pick a random binary task (2, 4, 5)
binary_task = random.choice([2, 4, 5])
data = (get_tsv_data(f"data/ARSC-Yu/raw/{label}.t{binary_task}.train",
label=label) +
get_tsv_data(f"data/ARSC-Yu/raw/{label}.t{binary_task}.dev",
label=label) +
get_tsv_data(f"data/ARSC-Yu/raw/{label}.t{binary_task}.test",
label=label))
random.shuffle(data)
task = collections.defaultdict(list)
for d in data:
task[d['label']].append(d['sentence'])
task = dict(task)
for key, val in task.items():
random.shuffle(val)
episode = {"xs": [list(task[k]) for k in task.keys()]}
return episode
def get_ARSC_test_tasks():
labels = sorted(set([line.strip() for line in open("data/ARSC-Yu/raw/workspace.target.list", "r").readlines()]))
tasks = list()
for label in labels:
for binary_task in (2, 4, 5):
train_data = get_tsv_data(f"data/ARSC-Yu/raw/{label}.t{binary_task}.train", label=label)
valid_data = get_tsv_data(f"data/ARSC-Yu/raw/{label}.t{binary_task}.dev", label=label)
test_data = get_tsv_data(f"data/ARSC-Yu/raw/{label}.t{binary_task}.test", label=label)
tasks.append({
"xs": [
[d['sentence'] for d in train_data if d['label'] == f"{label}-1"],
[d['sentence'] for d in train_data if d['label'] == f"{label}1"],
],
"x_valid": [
[d['sentence'] for d in valid_data if d['label'] == f"{label}-1"],
[d['sentence'] for d in valid_data if d['label'] == f"{label}1"],
],
"x_test": [
[d['sentence'] for d in test_data if d['label'] == f"{label}-1"],
[d['sentence'] for d in test_data if d['label'] == f"{label}1"],
],
})
assert all([len(task['xs'][0]) == len(task['xs'][1]) for task in tasks])
return tasks
def create_ARSC_train_episode(prefix: str = "data/ARSC-Yu/raw",
n_support: int = 5,
n_query: int = 5,
n_unlabeled=0):
labels = sorted(
set([
line.strip() for line in open(f"{prefix}/workspace.filtered.list",
"r").readlines()
]) - set([
line.strip() for line in open(f"{prefix}/workspace.target.list",
"r").readlines()
]))
# Pick a random label
label = random.choice(labels)
# Pick a random binary task (2, 4, 5)
binary_task = random.choice([2, 4, 5])
# Fix: this label/binary task sucks
while label == "office_products" and binary_task == 2:
# Pick a random label
label = random.choice(labels)
# Pick a random binary task (2, 4, 5)
binary_task = random.choice([2, 4, 5])
data = (
get_tsv_data(f"{prefix}/{label}.t{binary_task}.train", label=label) +
get_tsv_data(f"{prefix}/{label}.t{binary_task}.dev", label=label) +
get_tsv_data(f"{prefix}/{label}.t{binary_task}.test", label=label))
random.shuffle(data)
task = collections.defaultdict(list)
for d in data:
task[d['label']].append(d['sentence'])
task = dict(task)
assert min([len(val) for val in task.values()]) >= n_support + n_query + n_unlabeled, \
f"Label {label}_{binary_task}: min samples is {min([len(val) for val in task.values()])} while K+Q+U={n_support + n_query + n_unlabeled}"
for key, val in task.items():
random.shuffle(val)
episode = {
"xs": [[task[k][i] for i in range(n_support)] for k in task.keys()],
"xq":
[[task[k][n_support + i] for i in range(n_query)] for k in task.keys()]
}
if n_unlabeled:
episode['xu'] = [
item for k in task.keys()
for item in task[k][n_support + n_query:n_support + n_query +
n_unlabeled]
]
return episode
def create_ARSC_test_episode(prefix: str = "data/ARSC-Yu/raw",
n_query: int = 5,
n_unlabeled=0,
set_type: str = "test"):
assert set_type in ("test", "dev")
labels = [
line.strip()
for line in open(f"{prefix}/workspace.target.list", "r").readlines()
]
# Pick a random label
label = random.choice(labels)
# Pick a random binary task (2, 4, 5)
binary_task = random.choice([2, 4, 5])
support_data = get_tsv_data(f"{prefix}/{label}.t{binary_task}.train",
label=label)
assert len(support_data) == 10 # 2 * 5 shots
support_dict = collections.defaultdict(list)
for d in support_data:
support_dict[d['label']].append(d['sentence'])
query_data = get_tsv_data(
f"data/ARSC-Yu/raw/{label}.t{binary_task}.{set_type}", label=label)
query_dict = collections.defaultdict(list)
for d in query_data:
query_dict[d['label']].append(d['sentence'])
assert min([len(val)
for val in query_dict.values()]) >= n_query + n_unlabeled
for key, val in query_dict.items():
random.shuffle(val)
episode = {
"xs": [[sentence for sentence in support_dict[k]]
for k in sorted(query_dict.keys())],
"xq": [[query_dict[k][i] for i in range(n_query)]
for k in sorted(query_dict.keys())]
}
if n_unlabeled:
episode['xu'] = [
item for k in sorted(query_dict.keys())
for item in query_dict[k][n_query:n_query + n_unlabeled]
]
return episode
import shutil
import os
from utils.data import get_tsv_data
os.makedirs("data/ARSC-fixed", exist_ok=True)
with open("data/ARSC-Yu/raw/workspace.filtered.list", "r") as file:
train_labels = [line.strip() for line in file.readlines()]
with open("data/ARSC-Yu/raw/workspace.target.list", "r") as file:
test_labels = [line.strip() for line in file.readlines()]
train_data = list()
for lab in train_labels:
for ix in (2, 4, 5):
for split_type in ("train", "dev", "test"):
train_data += get_tsv_data(
f"data/ARSC-Yu/raw/{lab}.t{ix}.{split_type}", label=lab)
test_data = list()
for lab in test_labels:
for ix in (2, 4, 5):
for split_type in ("train", "dev", "test"):
test_data += get_tsv_data(
f"data/ARSC-Yu/raw/{lab}.t{ix}.{split_type}", label=lab)