eegpp3-beta 1.0__tar.gz

eegpp3_beta-1.0/LICENSE

@@ -0,0 +1 @@
+(C) Tuan Vu from DSLab, SoICT, HUST

eegpp3_beta-1.0/PKG-INFO

@@ -0,0 +1,69 @@
+Metadata-Version: 2.4
+Name: eegpp3-beta
+Version: 1.0
+Summary: EEG Phrase Predictor ver 3
+Author-email: "Duc Anh Nguyen, Vu An Tuan" <vuanhtuan1407@gmail.com>
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: dropbox>=12.0.2
+Requires-Dist: joblib>=1.4.2
+Requires-Dist: numpy>=2.1.1
+Requires-Dist: tqdm>=4.66.5
+Requires-Dist: torchinfo>=1.8.0
+Requires-Dist: scipy>=1.14.1
+Requires-Dist: pyyaml>=6.0.2
+Requires-Dist: pywavelets>=1.7.0
+Requires-Dist: ptwt>=0.1.9
+Requires-Dist: pandas>=2.2.2
+Requires-Dist: lightning>=2.4.0
+Requires-Dist: scikit-learn>=1.5.2
+Requires-Dist: seaborn>=0.13.2
+Dynamic: license-file
+
+# EEG Phase Predictor ver 2
+
+**Note: This is beta version, use for training with default dataset and inference only
+
+## Setup
+
+- Requirements: python >= 3.10
+- Installing:
+
+```aiignore
+  pip install eegpp3-beta
+```
+
+## Train with default dataset
+
+```aiignore
+    python -m eegpp3 --mode "train" --model_type <mode_type> --lr <learning_rate> --batch_size <batch_size> --n_epochs <num_epochs> --n_splits <num_folds> --resume_checkpoint <resume_from_checkpoint>
+```
+
+ex: `python -m eegpp3 --mode "train" --model_type "stftcnn1dnc" --n_epochs 20 --n_splits 10 --resume_checkpoint False`
+
+## Inference
+
+1. Old version (**Recommend**)
+
+```aiignore
+    python -m eegpp3 -p <path_to_infer_config.yml>
+```
+
+ex: `python -m eegpp3 -p "data_config_infer.yml"`
+
+2. New version
+
+```aiignore
+  python -m eegpp3 --mode "infer" --data_path <path_to_data_file> --infer_path <path_to_saving_file> --model_type <model_type>
+```
+
+ex:
+`python -m eegpp3 --mode "infer" --data_path "./dump_eeg_1.pkl" --infer_path './inference_result.txt" --model_type "stftcnn1dnc"`
+
+## Model type
+
+- stftcnn1dnc: Multi-channels STFT-CNN

eegpp3_beta-1.0/README.md

@@ -0,0 +1,43 @@
+# EEG Phase Predictor ver 2
+
+**Note: This is beta version, use for training with default dataset and inference only
+
+## Setup
+
+- Requirements: python >= 3.10
+- Installing:
+
+```aiignore
+  pip install eegpp3-beta
+```
+
+## Train with default dataset
+
+```aiignore
+    python -m eegpp3 --mode "train" --model_type <mode_type> --lr <learning_rate> --batch_size <batch_size> --n_epochs <num_epochs> --n_splits <num_folds> --resume_checkpoint <resume_from_checkpoint>
+```
+
+ex: `python -m eegpp3 --mode "train" --model_type "stftcnn1dnc" --n_epochs 20 --n_splits 10 --resume_checkpoint False`
+
+## Inference
+
+1. Old version (**Recommend**)
+
+```aiignore
+    python -m eegpp3 -p <path_to_infer_config.yml>
+```
+
+ex: `python -m eegpp3 -p "data_config_infer.yml"`
+
+2. New version
+
+```aiignore
+  python -m eegpp3 --mode "infer" --data_path <path_to_data_file> --infer_path <path_to_saving_file> --model_type <model_type>
+```
+
+ex:
+`python -m eegpp3 --mode "infer" --data_path "./dump_eeg_1.pkl" --infer_path './inference_result.txt" --model_type "stftcnn1dnc"`
+
+## Model type
+
+- stftcnn1dnc: Multi-channels STFT-CNN

eegpp3_beta-1.0/pyproject.toml

@@ -0,0 +1,38 @@
+[build-system]
+requires = ["setuptools>=61.0"]
+build-backend = "setuptools.build_meta"
+[project]
+name = "eegpp3-beta"
+version = "1.0"
+authors = [
+    { name = "Duc Anh Nguyen, Vu An Tuan", email = "vuanhtuan1407@gmail.com" },
+]
+description = "EEG Phrase Predictor ver 3"
+readme = "README.md"
+requires-python = ">=3.10"
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "License :: OSI Approved :: MIT License",
+    "Operating System :: OS Independent",
+]
+
+dependencies = [
+    'dropbox>=12.0.2',
+    'joblib>=1.4.2',
+    'numpy>=2.1.1',
+    'tqdm>=4.66.5',
+    'torchinfo>=1.8.0',
+    'scipy>=1.14.1',
+    'pyyaml>=6.0.2',
+    'pywavelets>=1.7.0',
+    'ptwt>=0.1.9',
+    'pandas>=2.2.2',
+    'lightning>=2.4.0',
+    'scikit-learn>=1.5.2',
+    'seaborn>=0.13.2',
+]
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+eegpp2 = ["configs/*.yml", "TOKEN.txt"]

eegpp3_beta-1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

eegpp3_beta-1.0/src/eegpp2/TOKEN.txt

@@ -0,0 +1,3 @@
+fq7ro54b8vcrz85
+i4foei4891bdnah
+yFJ0pz4d4TcAAAAAAAAAAV38Cd7Q1DXwn4L5dvu2K5zohoAWpG8mQqiNkbcKUTfp

eegpp3_beta-1.0/src/eegpp2/__init__.py

@@ -0,0 +1,6 @@
+import os
+
+EEGPP_DIR = os.path.dirname(os.path.abspath(__file__))
+
+CACHE_DIR = os.path.join(EEGPP_DIR, 'cache')
+os.makedirs(CACHE_DIR, exist_ok=True)

eegpp3_beta-1.0/src/eegpp2/__main__.py

@@ -0,0 +1,98 @@
+import argparse
+
+from . import params
+from .inference import infer, infer2
+from .trainer import EEGKFoldTrainer
+from .visualization import visualize_results
+
+
+def parse_arguments():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--mode", type=str, default='infer', choices=['infer', 'train'])
+    parser.add_argument("--model_type", type=str, default="stftcnn1dnc")
+    parser.add_argument("--lr", type=float, default=5e-4)
+    parser.add_argument("--batch_size", type=int, default=10)
+    parser.add_argument("--n_epochs", type=int, default=2)
+    parser.add_argument("--n_splits", type=int, default=2)
+    parser.add_argument("--n_workers", type=int, default=0)
+    parser.add_argument("--resume_checkpoint", type=bool, default=False)
+    parser.add_argument("--checkpoint_path", type=str, default=None)
+    parser.add_argument('--auto_visualize', type=bool, default=True)
+    parser.add_argument("--early_stopping", type=int, default=None)
+    parser.add_argument("--export_torchscript", type=bool, default=False)
+    parser.add_argument('--data_file', type=str, default='default',
+                        help='Need to specify data file path when in infer mode')
+    parser.add_argument("--infer_path", type=str, default=None)
+    parser.add_argument("--remove_tmp", type=bool, default=True)
+    parser.add_argument('--accelerator', type=str, default='auto')
+    parser.add_argument("-p", "--yaml_config_path", type=str, default='../../data_config_infer2.yml')
+
+    return parser.parse_args()
+
+#
+# def parse_options():
+#     parser = argparse.ArgumentParser()
+#     parser.add_argument('--data_file', type=str, default='default')
+#     parser.add_argument("--infer_path", type=str, default=None)
+#     parser.add_argument("--model_type", type=str, default="stftcnn1dnc")
+#     parser.add_argument("--batch_size", type=int, default=10)
+#     parser.add_argument("--n_workers", type=int, default=0)
+#     parser.add_argument("--checkpoint_path", type=str, default=None)
+#     return parser.parse_args()
+#
+
+def run():
+    args = parse_arguments()
+    if args.mode == 'train':
+        trainer = EEGKFoldTrainer(
+            model_type=args.model_type,
+            lr=args.lr,
+            batch_size=args.batch_size,
+            n_splits=args.n_splits,
+            n_epochs=args.n_epochs,
+            n_workers=args.n_workers,
+            accelerator=args.accelerator,
+            devices=params.DEVICES,
+            early_stopping=None,  # Force not apply early stopping because of KFold training process
+            export_torchscript=args.export_torchscript,
+            resume_checkpoint=args.resume_checkpoint,
+            checkpoint_path=args.checkpoint_path,
+        )
+        trainer.fit()
+        trainer.test()
+
+        if args.auto_visualize:
+            visualize_results(args.model_type)
+    else:
+        # opts = parse_options()
+        opts = args  # dummy code
+        if args.data_file == 'default' and opts.yaml_config_path is None:
+            raise ValueError('--data_file must be specified in infer mode')
+        else:
+            if opts.yaml_config_path is not None:
+                params.DATA_CONFIG_PATH = opts.yaml_config_path
+                infer2(opts)
+
+            else:
+                infer(
+                    data_path=opts.data_file,
+                    infer_path=opts.infer_path,
+                    model_type=opts.model_type,
+                    batch_size=opts.batch_size,
+                    n_workers=opts.n_workers,
+                    checkpoint_path=opts.checkpoint_path,
+                    remove_tmp=opts.remove_tmp,
+                )
+
+
+if __name__ == "__main__":
+    run()
+
+    # import torch
+    #
+    # t = torch.rand(2, 5120)
+    # window = torch.hamming_window(2048)
+    # rf = torch.fft.rfft(t)
+    # rs = torch.stft(t, n_fft=2048, win_length=2048, hop_length=512, normalized=True, return_complex=True, window=window)
+    # rt = torch.sqrt(rs.real ** 2 + rs.imag ** 2)
+    # print(rt, rt.shape)

eegpp3_beta-1.0/src/eegpp2/configs/__init__.py

@@ -0,0 +1,7 @@
+import os
+
+from .. import EEGPP_DIR
+
+CONFIG_DIR = os.path.join(EEGPP_DIR, 'configs')
+
+tmp = os.path.abspath(os.path.dirname(__file__))

eegpp3_beta-1.0/src/eegpp2/configs/cnn1d_config.yml

@@ -0,0 +1,20 @@
+turnoff: null
+conv_layers:
+  - name: conv1
+    out_channels: 4
+    kernel_size: 3
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 2
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv2
+    out_channels: 3
+    kernel_size: 3
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 2
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1

eegpp3_beta-1.0/src/eegpp2/configs/fftcnn1dnc_config.yml

@@ -0,0 +1,48 @@
+num_chains: 3
+n_fft: 5120
+conv_layers:
+  - name: conv1
+    out_channels: 4096
+    kernel_size: 11
+    stride: 4
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv2
+    out_channels: 5120
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv3
+    out_channels: 6144
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv4
+    out_channels: 3072
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv5
+    out_channels: 1024
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1

eegpp3_beta-1.0/src/eegpp2/configs/ffttransnc_config.yml

@@ -0,0 +1,7 @@
+num_chains: 3
+n_fft: 1024
+d_model: 1024
+dropout: 0.1
+nhead: 8
+dim_feedforward: 2048
+num_layers: 1

eegpp3_beta-1.0/src/eegpp2/configs/stftcnn1dnc_config.yml

@@ -0,0 +1,24 @@
+num_chains: 3
+n_fft: 2048
+win_length: 2048
+hop_length: 512
+normalized: True
+conv_layers:
+  - name: conv1
+    out_channels: 4096
+    kernel_size: 3
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv2
+    out_channels: 1024
+    kernel_size: 1
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 2
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1

eegpp3_beta-1.0/src/eegpp2/configs/stfttransnc_config.yml

@@ -0,0 +1,10 @@
+num_chains: 3
+n_fft: 1024
+win_length: 1024
+hop_length: 512
+return_complex: True
+normalized: True
+dropout: 0.1
+nhead: 4
+dim_feedforward: 2048
+num_layers: 1

eegpp3_beta-1.0/src/eegpp2/configs/transformer_config.yml

@@ -0,0 +1,7 @@
+d_model: 512
+dropout: 0.1
+nhead_encoder: 4
+nhead_decoder: 4
+num_encoder_layers: 1
+num_decoder_layers: 1
+dim_feedforward: 2048

eegpp3_beta-1.0/src/eegpp2/configs/wtcnn1dnc_config.yml

@@ -0,0 +1,49 @@
+num_chains: 3
+emb_size: 100
+wavelet: "morl"
+conv_layers:
+  - name: conv1
+    out_channels: 4096
+    kernel_size: 11
+    stride: 4
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv2
+    out_channels: 5120
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv3
+    out_channels: 6144
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv4
+    out_channels: 3072
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1
+  - name: conv5
+    out_channels: 1024
+    kernel_size: 5
+    stride: 1
+    padding: 0
+    pooling_kernel_size: 3
+    pooling_stride: 2
+    pooling_padding: 0
+    dropout: 0.1

eegpp3_beta-1.0/src/eegpp2/configs/wtresnet501dnc_config.yml

@@ -0,0 +1,3 @@
+num_chains: 3
+emb_size: 100
+wavelet: "morl"

eegpp3_beta-1.0/src/eegpp2/data/__init__.py

@@ -0,0 +1,39 @@
+import os
+from pathlib import Path
+
+DATA_DIR = os.path.abspath(os.path.dirname(__file__))
+RAW_DATA_DIR = os.path.join(DATA_DIR, 'raw')
+DUMP_DATA_DIR = os.path.join(DATA_DIR, 'dump')
+
+os.makedirs(DUMP_DATA_DIR, exist_ok=True)
+
+SEQ_FILES = [
+    str(Path(RAW_DATA_DIR, "raw_K3_EEG3_11h.txt")),
+    str(Path(RAW_DATA_DIR, "raw_RS2_EEG1_23 hr.txt")),
+    #str(Path(RAW_DATA_DIR, "raw_S1_EEG1_23 hr.txt")),
+    #str(Path(RAW_DATA_DIR, "K1_EEG1_SAL.csv")),
+    #str(Path(RAW_DATA_DIR, "K1_EEG7_SAL.csv")),
+    #str(Path(RAW_DATA_DIR, "K2_EEG4_SAL.csv")),
+    #str(Path(RAW_DATA_DIR, "K2_EEG5_SAL.csv")),
+    #str(Path(RAW_DATA_DIR, "K4_EEG7_SAL.csv")),
+]
+
+LABEL_FILES = [
+    str(Path(RAW_DATA_DIR, "K3_EEG3_11h.txt")),
+    str(Path(RAW_DATA_DIR, "RS2_EEG1_23 hr.txt")),
+    #str(Path(RAW_DATA_DIR, "S1_EEG1_23 hr.txt")),
+    #str(Path(RAW_DATA_DIR, "K1_EEG1_11h.txt")),
+    #str(Path(RAW_DATA_DIR, "K1_EEG7_11h.txt")),
+    #str(Path(RAW_DATA_DIR, "K2_EEG4_11h.txt")),
+    #str(Path(RAW_DATA_DIR, "K2_EEG5_11h.txt")),
+    #str(Path(RAW_DATA_DIR, "K4_EEG7_11h.txt")),
+]
+
+DUMP_DATA_FILES = {
+    "train": [
+        str(Path(DUMP_DATA_DIR, f"dump_eeg_{i+1}.pkl")) for i in range(len(SEQ_FILES))
+    ],
+    "infer": [
+        str(Path(DUMP_DATA_DIR, f"dump_eeg_{i+1}_infer.pkl")) for i in range(len(SEQ_FILES))
+    ],
+}

eegpp3_beta-1.0/src/eegpp2/dataloader.py

@@ -0,0 +1,177 @@
+import os
+
+import numpy as np
+import torch
+from sklearn.model_selection import KFold
+from torch.utils.data import Subset, ConcatDataset, DataLoader, random_split, Sampler, RandomSampler
+
+from .data import DUMP_DATA_FILES, DUMP_DATA_DIR
+from . import params
+from .dataset import EEGDataset
+from .utils.data_utils import get_dataset_train
+
+LABEL_MARKER = "EpochNo"
+
+def fread_header_labels(inp):
+    global SEP_CHECKED, SEPERATOR
+    fin = open(inp, errors='ignore')
+
+    headers = []
+
+    while True:
+        line = fin.readline()
+        if line == "":
+            break
+        headers.append(line)
+        if line.startswith(LABEL_MARKER):
+            break
+
+    return "".join(headers), fin
+
+class EEGKFoldSampler(Sampler):
+    def __init__(self, dataset, k_fold):
+        super().__init__()
+
+    def __iter__(self):
+        pass
+
+    def __len__(self):
+        pass
+
+
+class EEGKFoldDataLoader:
+    def __init__(
+            self,
+            dataset_files='default',
+            n_splits=5,
+            n_workers=0,
+            batch_size=4,
+            minmax_normalized=True,
+    ):
+        """
+        :param dataset_files: must be list of Path to dump file or "default"
+        :param n_splits:
+        :param n_workers:
+        :param batch_size:
+        :param minmax_normalized:
+        """
+
+        self.minmax_normalized = minmax_normalized
+
+        self.val_dataset = None
+        self.train_dataset = None
+        self.test_dataset = None
+        self.datasets = None
+        self.all_splits = None
+        self.train_val_datasets = None
+        self.current_fold = None
+        self.current_epoch = None
+
+        if isinstance(dataset_files, list):
+            self.dataset_files = dataset_files
+        else:
+            self.prepare_default_data()
+            self.dataset_files = DUMP_DATA_FILES['train']
+
+        self.n_splits = n_splits
+        self.n_workers = n_workers
+        self.batch_size = batch_size
+        self.split_generator = torch.Generator().manual_seed(params.RD_SEED)
+        self.dataloader_generator = torch.Generator().manual_seed(0)
+        self.setup()
+
+    @staticmethod
+    def prepare_default_data():
+        os.makedirs(DUMP_DATA_DIR, exist_ok=True)
+        for _, _, files in os.walk(DUMP_DATA_DIR):
+            if len(files) != len(DUMP_DATA_FILES['train']):
+                get_dataset_train(remote_type='dump')
+
+    def setup(self):
+        print("Loading data...")
+        datasets = []
+        train_val_dts, test_dts = [], []
+        for i, _ in enumerate(self.dataset_files):
+            # print(type(DUMP_DATA_FILES), DUMP_DATA_FILES)
+            dump_file = DUMP_DATA_FILES['train'][i]
+            print("Loading dump file {}".format(dump_file))
+            i_dataset = EEGDataset(dump_file, w_out=params.W_OUT, minmax_normalized=self.minmax_normalized)
+            datasets.append(i_dataset)
+            train_val_dt, test_dt = random_split(i_dataset, [0.9, 0.1], generator=self.split_generator)
+            train_val_dts.append(train_val_dt)
+            test_dts.append(test_dt)
+
+        self.datasets = datasets
+        self.train_val_datasets = train_val_dts
+        self.test_dataset = ConcatDataset(test_dts)
+        kf = KFold(n_splits=self.n_splits, shuffle=True, random_state=params.RD_SEED)
+        all_splits = []
+        for i, _ in enumerate(self.datasets):
+            splits = [subset for subset in kf.split(train_val_dts[i])]
+            all_splits.append(splits)
+        self.all_splits = all_splits
+
+    def set_fold(self, k):
+        if k is None or k < 0 or k >= self.n_splits:
+            print("Fold value is invalid. Set to default value: 0")
+            self.current_fold = 0
+        else:
+            self.current_fold = k
+
+        train_dts, val_dts = [], []
+        for i, (dataset, splits) in enumerate(zip(self.datasets, self.all_splits)):
+            train_ids, val_ids = splits[k]
+            train_subset_ids = np.array(self.train_val_datasets[i].indices)[train_ids]
+            val_subset_ids = np.array(self.train_val_datasets[i].indices)[val_ids]
+            train_dt = Subset(dataset, train_subset_ids)
+            val_dt = Subset(dataset, val_subset_ids)
+            train_dts.append(train_dt)
+            val_dts.append(val_dt)
+
+        self.train_dataset = ConcatDataset(train_dts)
+        self.val_dataset = ConcatDataset(val_dts)
+
+    def set_epoch(self, epoch):
+        if self.current_fold is None:
+            print("Fold attribute is None. Set to default value: 0")
+            self.current_fold = 0
+        self.current_epoch = epoch
+        dataloader_generator_seed = (self.current_epoch + 1) * (self.current_fold + 1)
+        self.dataloader_generator = self.dataloader_generator.manual_seed(dataloader_generator_seed)
+
+    def train_dataloader(self, epoch):
+        self.set_epoch(epoch)
+        return DataLoader(
+            dataset=self.train_dataset,
+            batch_size=self.batch_size,
+            shuffle=False,
+            sampler=RandomSampler(self.train_dataset, generator=self.dataloader_generator),
+            num_workers=self.n_workers,
+            drop_last=True
+        )
+
+    def val_dataloader(self):
+        return DataLoader(
+            dataset=self.val_dataset,
+            batch_size=self.batch_size,
+            shuffle=False,
+            num_workers=self.n_workers,
+            drop_last=True,
+        )
+
+    def test_dataloader(self):
+        return DataLoader(
+            dataset=self.test_dataset,
+            batch_size=self.batch_size,
+            shuffle=False,
+            num_workers=self.n_workers,
+            drop_last=True,
+        )
+
+
+if __name__ == '__main__':
+    dataloader = EEGKFoldDataLoader()
+    dataloader.set_fold(0)
+    train_dataloader = dataloader.train_dataloader(epoch=0)
+    for i, data in enumerate(train_dataloader):
+        print(data)