"""
The setting of Superb ER
Authors
* Tzu-Hsien Huang 2021
* Leo 2021
* Leo 2022
"""
import json
import logging
from pathlib import Path
from typing import List
import pandas as pd
import torch
from omegaconf import MISSING
from torch.utils.data import random_split
from s3prl.dataio.corpus.iemocap import IEMOCAP
from s3prl.util.download import download
from .superb_sid import SuperbSID
logger = logging.getLogger(__name__)
__all__ = [
"iemocap_for_superb",
"SuperbER",
]
[docs]def iemocap_for_superb(
target_dir: str,
cache_dir: str,
iemocap: str,
test_fold: int,
valid_ratio: float = 0.2,
get_path_only: bool = False,
):
"""
Prepare IEMOCAP for emotion classfication with SUPERB protocol,
following :obj:`SuperbER.prepare_data` format.
.. note::
In SUPERB protocol, you need to do 5-fold cross validation.
Also, only use 4 emotion classes: :code:`happy`, :code:`angry`,
:code:`neutral`, and :code:`sad` with balanced data points and
the :code:`excited` class is merged into :code:`happy` class.
Args:
iemocap (str): The root path of the IEMOCAP
test_fold (int): Which fold to use as the test fold, select from 0 to 4
valid_ratio (float): given the remaining 4 folds, how many data to use as the validation set
**others: refer to :obj:`SuperbER.prepare_data`
"""
target_dir = Path(target_dir)
train_path = target_dir / f"train.csv"
valid_path = target_dir / f"valid.csv"
test_paths = [target_dir / f"test.csv"]
if get_path_only:
return train_path, valid_path, test_paths
corpus = IEMOCAP(iemocap)
all_datapoints = corpus.all_data
def format_fields(data: dict):
result = dict()
for data_id in data.keys():
datapoint = data[data_id]
result[data_id] = dict(
wav_path=datapoint["wav_path"],
label=datapoint["emotion"],
)
return result
def filter_data(data_ids: List[str]):
result = dict()
for data_id in data_ids:
data_point = all_datapoints[data_id]
if data_point["emotion"] in ["neu", "hap", "ang", "sad", "exc"]:
if data_point["emotion"] == "exc":
data_point["emotion"] = "hap"
result[data_id] = data_point
return result
test_session_id = test_fold + 1
train_meta_data_json = (
Path(cache_dir) / f"test_session{test_session_id}_train_metadata.json"
)
test_meta_data_json = (
Path(cache_dir) / f"test_session{test_session_id}_test_metadata.json"
)
download(
train_meta_data_json,
f"https://huggingface.co/datasets/s3prl/iemocap_split/raw/4097f2b496c41eed016d4e5eb0ada4cccd46d1f3/Session{test_session_id}/train_meta_data.json",
refresh=False,
)
download(
test_meta_data_json,
f"https://huggingface.co/datasets/s3prl/iemocap_split/raw/4097f2b496c41eed016d4e5eb0ada4cccd46d1f3/Session{test_session_id}/test_meta_data.json",
refresh=False,
)
with open(train_meta_data_json) as f:
metadata = json.load(f)["meta_data"]
dev_ids = [Path(item["path"]).stem for item in metadata]
with open(test_meta_data_json) as f:
metadata = json.load(f)["meta_data"]
test_ids = [Path(item["path"]).stem for item in metadata]
train_len = int((1 - valid_ratio) * len(dev_ids))
train_valid_lens = [train_len, len(dev_ids) - train_len]
torch.manual_seed(0)
train_ids, valid_ids = random_split(dev_ids, train_valid_lens)
train_data = format_fields(filter_data(train_ids))
valid_data = format_fields(filter_data(valid_ids))
test_data = format_fields(filter_data(test_ids))
def dict_to_csv(data_dict, csv_path):
keys = sorted(list(data_dict.keys()))
fields = sorted(data_dict[keys[0]].keys())
data = dict()
for field in fields:
data[field] = []
for key in keys:
data[field].append(data_dict[key][field])
data["id"] = keys
df = pd.DataFrame(data)
df.to_csv(csv_path, index=False)
dict_to_csv(train_data, train_path)
dict_to_csv(valid_data, valid_path)
dict_to_csv(test_data, test_paths[0])
return train_path, valid_path, test_paths
[docs]class SuperbER(SuperbSID):
[docs] def default_config(self) -> dict:
return dict(
start=0,
stop=None,
target_dir=MISSING,
cache_dir=None,
remove_all_cache=False,
prepare_data=dict(
iemocap=MISSING,
test_fold=MISSING,
),
build_encoder=dict(),
build_dataset=dict(),
build_batch_sampler=dict(
train=dict(
batch_size=4,
shuffle=True,
),
valid=dict(
batch_size=4,
),
test=dict(
batch_size=4,
),
),
build_upstream=dict(
name=MISSING,
),
build_featurizer=dict(
layer_selections=None,
normalize=False,
),
build_downstream=dict(
hidden_size=256,
),
build_model=dict(
upstream_trainable=False,
),
build_task=dict(),
build_optimizer=dict(
name="Adam",
conf=dict(
lr=1.0e-4,
),
),
build_scheduler=dict(
name="ExponentialLR",
gamma=0.9,
),
save_model=dict(),
save_task=dict(),
train=dict(
total_steps=30000,
log_step=500,
eval_step=1000,
save_step=1000,
gradient_clipping=1.0,
gradient_accumulate=8,
valid_metric="accuracy",
valid_higher_better=True,
auto_resume=True,
resume_ckpt_dir=None,
),
evaluate=dict(),
)
[docs] def prepare_data(
self,
prepare_data: dict,
target_dir: str,
cache_dir: str,
get_path_only: bool = False,
):
"""
Prepare the task-specific data metadata (path, labels...).
By default call :obj:`iemocap_for_superb` with :code:`**prepare_data`
Args:
prepare_data (dict): same in :obj:`default_config`,
support arguments in :obj:`iemocap_for_superb`
target_dir (str): Parse your corpus and save the csv file into this directory
cache_dir (str): If the parsing or preprocessing takes too long time, you can save
the temporary files into this directory. This directory is expected to be shared
across different training sessions (different hypers and :code:`target_dir`)
get_path_only (str): Directly return the filepaths no matter they exist or not.
Returns:
tuple
1. train_path (str)
2. valid_path (str)
3. test_paths (List[str])
Each path (str) should be a csv file containing the following columns:
==================== ====================
column description
==================== ====================
id (str) - the unique id for this data point
wav_path (str) - the absolute path of the waveform file
label (str) - a string label of the waveform
start_sec (float) - optional, load the waveform from :code:`start_sec` seconds. If not presented or is :code:`math.nan`, load from the beginning.
end_sec (float) - optional, load the waveform from :code:`end_sec` seconds. If not presented or is :code:`math.nan`, load to the end.
==================== ====================
"""
return iemocap_for_superb(
**self._get_current_arguments(flatten_dict="prepare_data")
)