Update docs and yaml for baselines

docs/replication-guide/baselines.md

@@ -1,2 +1,46 @@
 # Reproduce Baselines
-:hourglass_flowing_sand: Coming soon! :hourglass_flowing_sand:
+
+Reproducing baselines used in this study is very similar to the adaptation of the FM as essentially we are just using the FM weights for adaptation. 
+
+## Randomly initialized
+
+We provide the YAML configuration to train the random init baseline at `experiments/baselines/supervised_training/supervised_random_init.yaml`
+
+By default, we configure this for Task 1. You can adapt this for Task 2 and Task 3 by searching for `Note: ` comments in the YAML that outline what must be changed. 
+
+You can start training by running this in the root code folder,
+```bash
+lighter fit --config_file ./experiments/baselines/supervised_training/supervised_random_init.yaml
+```
+
+## Transfer learning
+We provide the YAML configuration to train the transfer learning baseline at `experiments/baselines/supervised_training/supervised_finetune.yaml`
+
+This baseline is only used for Task 2 and Task 3 as we use the random init baseline from Task 1 for the transfer. Follow the `Note: ` comments to switch between Task 2 and Task 3 configurations. 
+
+You can start training by running this in the root code folder,
+```bash
+lighter fit --config_file ./experiments/baselines/supervised_training/supervised_finetune.yaml
+```
+
+## Med3D / MedicalNet
+Original repo: https://github.com/Tencent/MedicalNet
+
+We have provided re-implementations of Med3D to fit into our YAML workflows at `experiments/baselines/med3d/finetune.yaml`. Again, the `Note: ` comments help adapt for different tasks. 
+
+
+You can start training by running this in the root code folder,
+```bash
+lighter fit --config_file ./experiments/baselines/med3d/finetune.yaml
+```
+
+## Models Genesis
+Original repo: https://github.com/MrGiovanni/ModelsGenesis
+
+We have provided re-implementations of Models Genesis to fit into our YAML workflows at `experiments/baselines/models_genesis/finetune.yaml`. Again, the `Note: ` comments help adapt for different tasks. 
+
+
+You can start training by running this in the root code folder,
+```bash
+lighter fit --config_file ./experiments/baselines/models_genesis/finetune.yaml
+```

docs/replication-guide/fm_adaptation.md

@@ -6,7 +6,7 @@ The FM was adapted by either fine-tuning all its weights or by freezing its weig
 
 We provide the YAML configuration for this at `experiments/adaptation/fmcib_finetune.yaml`.
 
-By default, we configure this for Task 1. You can adapt this for Task 2 and Task 3 by searching for 'Note: ' comments in the YAML that outline what must be changed. Make sure you download the weights for the pre-trained foundation model before attempting to reproduce this training. 
+By default, we configure this for Task 1. You can adapt this for Task 2 and Task 3 by searching for `Note: ` comments in the YAML that outline what must be changed. Make sure you download the weights for the pre-trained foundation model before attempting to reproduce this training. 
 
 
 You can start training by running this in the root code folder,

docs/replication-guide/inference.md

@@ -12,10 +12,10 @@ lighter predict --config_file ./experiments/inference/extract_features.yaml
 !!! note
      While the above pipeline will allow you to extract features, we provide an easier and simpler, recommended API to do this. Please refer to [Quick Start](../getting-started/quick-start.md) or [Cloud Quick Start](../getting-started/cloud-quick-start.md)
 
-However, this method might be preferred when features need to be extracted from different models (used as baselines in our study). Follow the `Note:` in the corresponding config file to change model paths. 
 
-## Running predictions from our supervised models (Finetuned FM/ Baselines)
+However, this method might be preferred when features need to be extracted from different models (used as baselines in our study). Follow the `Note:` in the corresponding config file to change model paths and use different baselines tested.
 
+## Running predictions from our supervised models (Finetuned FM/ Baselines)
 
 To run predictions from our models (both supervised and self-supervised), we provide YAML files that can be run with the lighter interface. These are found in `experiments/inference`, namely `get_predictions.yaml` for getting the predictions. 
 
@@ -37,3 +37,5 @@ lighter predict --config_file ./experiments/inference/get_predictions.yaml
 ```
 As with the previous YAMLS, please follow the 'Note:' tags to place appropriate data paths and change relevant parameters. This YAML is to be used if you want to get target predictions from the models.
 
+!!! note
+     The predictions can be extracted for different tasks as well as different baselines by following the `Note:` comments. 

experiments/baselines/med3d/supervised_training/finetune.yaml → experiments/baselines/med3d/finetune.yaml

@@ -57,34 +57,32 @@ system:
     criterion:
         _target_: torch.nn.CrossEntropyLoss # Note: Change to torch.nn.BCEWithLogitsLoss for Task 2 and Task 3
 
-    optimizers:
+    optimizer:
         _target_: torch.optim.Adam
         params: "$@system#model.parameters()"
         # lr: 0.001 # Compute L  R dynamically for different batch sizes
         # weight_decay: 0.0
         # momentum: 0.9
 
-    schedulers:
-      scheduler:
+    scheduler:
         _target_: torch.optim.lr_scheduler.StepLR
         optimizer: "@system#optimizers"
         step_size: 30
-      strict: True
 
-    train_metrics: 
+    metrics: 
+      train:
         - _target_: torchmetrics.AveragePrecision
           task: binary # Note: Change to `binary` for Task 2 and Task 3 and remove num_classes below
           # num_classes: 8 
 
         - _target_: torchmetrics.AUROC
           task: binary # Note: Change to `binary` for Task 2 and Task 3 and remove num_classes below
           # num_classes: 8 
+      val: "%#train"
+      test: "%#train"
 
-
-    val_metrics: "%system#train_metrics"
-    test_metrics: "%system#train_metrics"
-
-    train_dataset:
+    datasets:
+      train:
         _target_: fmcib.datasets.SSLRadiomicsDataset
         path: null # Note: Change path
         label: "survival" # Note: Change to "malignancy" for Task 2 and "survival" for Task 3
@@ -118,7 +116,7 @@ system:
                 #   subtrahend: -1024
                 #   divisor: 3072
 
-    val_dataset:
+      val:
         _target_: fmcib.datasets.SSLRadiomicsDataset
         path: null # Note: Change path
         label: "@system#train_dataset#label"
@@ -141,6 +139,4 @@ system:
                 - _target_: monai.transforms.SpatialPad
                   spatial_size: [50, 50, 50]
 
-
-    test_dataset: null
-    predict_dataset: null
+      test: null