adding efficientnet weights + config

README.md

@@ -0,0 +1,15 @@
+## DeepECG-Tok_EfficientNetV2_77_Classes
+
+- DeepECG-Tok_EfficientNetV2_77_Classes is a repository for the EfficientNetV2 weights that classify embeddings from DeepECG-Tok into 77 classes.
+- base_config.yaml file contains the parameters for the training and inference.
+
+- Paths to be updated:
+    - base_checkpoint_path: path to the checkpoints directory
+    - pretrained_tokenizer_path: path to the pretrained tokenizer checkpoint
+    - inference_dataset_path: path to the inference dataset
+    - inference_checkpoint_path: path to the inference checkpoint
+
+- To run inference, run the following command:
+```bash
+bash scripts/runner.sh --base_config config/linear_probing/base_config.yaml --selected_gpus 0,1 --use_wandb true --run_mode inference
+```

base_config.yaml

@@ -0,0 +1,80 @@
+# pipeline project
+pipeline_project: !!str "ECG_Tokenizer_Linear_Probing"
+model_name: !!str "ECG_Tokenizer_Wrapper"
+runner_name: !!str "ECG_Tokenizer_Linear_Probing_Runner"
+run_mode: !!str "inference"
+num_epochs: !!int 10
+seed: !!int 42
+base_checkpoint_path: !!str "/volume/ECG_tokenizer/outputs/"
+pretrained_tokenizer_path: !!str "/volume/ECG_tokenizer/checkpoints/ECG_tokenizer_latest/best_model_epoch_10.pt"
+inference_dataset_path: !!str "/volume/ECG_tokenizer/outputs/preprocessed_bert_output.parquet"
+inference_checkpoint_path: !!str "/volume/ECG_tokenizer/checkpoints/ECG_Tokenizer_Linear_Probing/ECG_Tokenizer_Linear_Probing/5zg01bx6_20250824-041452/checkpoint_epoch_10.pt"
+# wandb parameters
+wandb_project: !!str "ECG_Tokenizer_Linear_Probing"
+wandb_entity: !!str "mhi_ai"
+use_wandb: !!bool false
+# training parameters
+lr: !!float 3e-4
+scheduler_type: !!str "cosine_warm_restart"
+lr_step_period: !!int 1
+factor: !!float 0.3
+optimizer: !!str "AdamW"
+weight_decay: !!float 1e-4
+step_size: !!int 1
+gamma: !!float 0.1
+num_warmup_percent: !!float 0.1
+num_hard_restarts_cycles: !!float 0.5
+warm_restart_tmult: !!int 2
+# VQVAE parameters
+decoder_mode: !!str "classification"
+decoder_name: !!str "EfficientNetV2_Classifier_Decoder"
+# Classification parameters
+num_classes: !!int 77
+criterion: !!str "bce_loss"
+# Dataset parameters
+train_dataset_path: !!str "/volume/ECG_tokenizer/output/MHI/mimic_mhi_psa_train_updated.parquet"
+validation_dataset_path: !!str "/volume/ECG_tokenizer/output/MHI/mhi_psa_test_updated.parquet"
+num_workers: !!int 32
+batch_size: !!int 12
+signal_path_column: !!str "ecg_path" # waveform_path_psa or waveform_path_original
+# waveform parameters
+waveform_length: !!int 2500
+num_leads: !!int 12
+normalize_waveforms: !!bool False
+lead_stats:
+  I:
+    mean: !!float -0.982577
+    std: !!float 33.551235
+  II:
+    mean: !!float -0.683182
+    std: !!float 35.616466
+  III:
+    mean: !!float 0.299394
+    std: !!float 40.365378
+  aVR:
+    mean: !!float 0.832880
+    std: !!float 28.102812
+  aVL:
+    mean: !!float -0.640986
+    std: !!float 32.563651
+  aVF:
+    mean: !!float -0.191894
+    std: !!float 34.169092
+  V1:
+    mean: !!float -1.351383
+    std: !!float 54.328896
+  V2:
+    mean: !!float -1.366718
+    std: !!float 72.522918
+  V3:
+    mean: !!float -1.325758
+    std: !!float 66.522440
+  V4:
+    mean: !!float -1.091183
+    std: !!float 61.623552
+  V5:
+    mean: !!float -0.937534
+    std: !!float 57.341500
+  V6:
+    mean: !!float -0.882349
+    std: !!float 52.247728

deepecg_tokenizer_efficientnet.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd9fed76ffc94d988dfe8611e5ce9fa12cc95f8ba9f0d227b4186db8f1f06d8e
+size 173070264