Init

Author: patsch007

.gitignore

@@ -0,0 +1,11 @@
+volume/*
+outputs/*
+Cache/*
+runs/*
+
+*.nii.gz
+*.raw
+*.mhd
+*.stl
+*cache*
+

Readme.md

@@ -0,0 +1 @@
+# CVSSL - Contrastive volumetric self supervised learning

callbacks/metric_evaluator.py

@@ -0,0 +1,89 @@
+# Copyright (c) 2025 lightning-hydra-boilerplate
+# Licensed under the MIT License.
+
+"""Callback for logging metrics during validation and test using PyTorch Lightning."""
+
+import torch
+from lightning.pytorch import Callback, LightningModule, Trainer
+from lightning.pytorch.utilities.types import STEP_OUTPUT
+
+
+class MetricEvaluator(Callback):
+    """Logs custom metrics during validation and test epochs.
+
+    Args:
+        metrics (Dict[str, List[torch.nn.Module]]): A dictionary mapping each stage
+            ("validation", "test") to a list of metric instances.
+
+    Example:
+            {
+                "validation": [Accuracy(...), F1Score(...)],
+                "test": [Accuracy(...), F1Score(...)]
+            }
+    """
+
+    def __init__(self, metrics: dict[str, list[torch.nn.Module]]) -> None:
+        self.metrics = metrics
+
+    def _reset(self, stage: str) -> None:
+        """Reset all metrics for a given stage."""
+        for metric in self.metrics.get(stage, []):
+            metric.reset()
+
+    def _update(self, stage: str, preds: torch.Tensor, targets: torch.Tensor) -> None:
+        """Update all metrics for a given stage with predictions and targets."""
+        for metric in self.metrics.get(stage, []):
+            metric.to(preds.device)
+            metric.update(preds, targets)
+
+    def _log(self, stage: str, pl_module: LightningModule) -> None:
+        """Compute and log all metrics for a given stage."""
+        # prefix = "val" if stage == "validation" else "test"
+        # for metric in self.metrics.get(stage, []):
+        #     name = metric.__class__.__name__.lower()
+        #     value = metric.compute()
+        #     pl_module.log(f"{prefix}_{name}", value, prog_bar=True)
+
+    def on_validation_epoch_start(self, _trainer: Trainer, _pl_module: LightningModule) -> None:
+        """Reset metrics at the start of validation epoch."""
+        self._reset("validation")
+
+    def on_validation_batch_end(
+        self,
+        _trainer: Trainer,
+        pl_module: LightningModule,
+        _outputs: STEP_OUTPUT,
+        batch: dict,
+        _batch_idx: int,
+        _dataloader_idx: int = 0,
+    ) -> None:
+        """Update validation metrics with each batch."""
+        x, y = batch[0], batch[1]
+        #preds = pl_module(x).argmax(dim=1)
+        #self._update("validation", preds, y)
+
+    def on_validation_epoch_end(self, _trainer: Trainer, pl_module: LightningModule) -> None:
+        """Compute and log validation metrics at end of epoch."""
+        self._log("validation", pl_module)
+
+    def on_test_epoch_start(self, _trainer: Trainer, _pl_module: LightningModule) -> None:
+        """Reset metrics at the start of test epoch."""
+        self._reset("test")
+
+    def on_test_batch_end(
+        self,
+        _trainer: Trainer,
+        pl_module: LightningModule,
+        _outputs: STEP_OUTPUT,
+        batch: dict,
+        _batch_idx: int,
+        _dataloader_idx: int = 0,
+    ) -> None:
+        """Update test metrics with each batch."""
+        # x, y = batch["image"], batch["label"]
+        # preds = pl_module(x).argmax(dim=1)
+        # self._update("test", preds, y)
+
+    def on_test_epoch_end(self, _trainer: Trainer, pl_module: LightningModule) -> None:
+        """Compute and log test metrics at end of epoch."""
+        self._log("test", pl_module)

configs/data/example_data.yaml

@@ -0,0 +1,4 @@
+_target_: data.example_data.lightning_datamodule.ExampleDataModule
+batch_size: 32
+num_workers: 4
+path: "/mnt/DATA/**/*.nii.gz"

configs/eval.yaml

@@ -0,0 +1,11 @@
+defaults:
+  - _self_
+  - model: ???
+  - data: ???
+  - trainer: default
+  - experiment: ???
+  - hydra: default
+
+mode: eval
+ckpt_path: ??? # User must override this
+data_split: "test" # one of: test or val

configs/experiment/example_experiment.yaml

@@ -0,0 +1,38 @@
+# @package _global_
+
+defaults:
+  - override /model: example_model # Override the model to use
+  - override /data: example_data # Override the data module to use
+  - override /trainer: base # Override or define trainer settings
+
+# Base configuration
+seed: 42
+skip_test: False
+experiment_name: "exp"
+
+# Model parameters
+model:
+  optimizer:
+    lr: 0.0005
+
+# Data parameters
+data:
+  batch_size: 18
+
+# Lightning Trainer parameters
+trainer:
+  max_epochs: 100
+  precision: 32
+  check_val_every_n_epoch: 4 
+  callbacks:
+    metric_evaluator:
+      _target_: callbacks.metric_evaluator.MetricEvaluator
+      metrics:
+        validation:
+          - _target_: torchmetrics.Accuracy
+            task: "multiclass"
+            num_classes: ${model.num_classes}
+        test:
+          - _target_: torchmetrics.Accuracy
+            task: "multiclass"
+            num_classes: ${model.num_classes}

configs/hydra/default.yaml

@@ -0,0 +1,19 @@
+defaults:
+  - override job_logging: colorlog
+  - override hydra_logging: colorlog
+
+run:
+  dir: "outputs/${mode}/${experiment_name}-${now:%Y-%m-%d_%H-%M-%S}"
+sweep:
+  dir: "outputs/sweeps/${mode}/${experiment_name}-${now:%Y-%m-%d_%H-%M-%S}"
+  subdir: ${hydra.job.override_dirname}
+job:
+  config:
+    override_dirname:
+      exclude_keys:
+        - experiment
+        - params_search
+job_logging:
+  handlers:
+    file:
+      filename: ${hydra.runtime.output_dir}/${mode}.log

configs/model/example_model.yaml

@@ -0,0 +1,14 @@
+_target_: model.example_model.lightning_module.ExampleLightningModel
+num_classes: 10
+optimizer:
+  _target_: torch.optim.Adam
+  _partial_: true
+  lr: 0.001
+loss_fn:
+  _target_: torch.nn.functional.cross_entropy
+  _partial_: true
+scheduler:
+  _target_: torch.optim.lr_scheduler.StepLR
+  step_size: 5
+  gamma: 0.1
+  _partial_: true

configs/params_search/example_optuna.yaml

@@ -0,0 +1,30 @@
+# @package _global_
+
+defaults:
+  - override /hydra/sweeper: optuna
+
+# Make sure this is aligned with the metric logged in lightning module
+optimized_metric: "val_multiclassaccuracy"
+
+# See: https://hydra.cc/docs/plugins/optuna_sweeper/
+hydra:
+  mode: "MULTIRUN"
+  sweeper:
+    _target_: hydra_plugins.hydra_optuna_sweeper.optuna_sweeper.OptunaSweeper
+    storage: null
+    study_name: null
+    n_jobs: 1
+    direction: maximize
+    n_trials: 3
+
+    #You can choose TPEsampler, RandomSampler, GridSampler, etc.
+    # See: https://optuna.readthedocs.io/en/stable/reference/samplers/index.html
+    sampler:
+      _target_: optuna.samplers.TPESampler
+      seed: 42
+      n_startup_trials: 3 # number of random sampling runs before optimization starts
+
+    # Hyperparameter search space
+    params:
+      model.optimizer.lr: interval(0.0001, 0.1)
+      data.batch_size: choice(32, 64)

configs/predict.yaml

@@ -0,0 +1,12 @@
+defaults:
+  - _self_
+  - model: ???
+  - data: ???
+  - trainer: default
+  - experiment: ???
+  - hydra: default
+
+mode: predict
+ckpt_path: ??? # User must override this
+data_split: "test" # one of: train, test, val, or predict
+save_format: csv # json or csv