restructure and bug correction

configs/config.yaml

@@ -35,7 +35,7 @@ defaults:
   - task: default.yaml
   #- predict: default.yaml
 
-  - experiment: MSAFFDebug.yaml  # default run is for debugging.
+  - experiment: SIMDebug.yaml  # default run is for debugging.
 
   - hydra: default.yaml
 

configs/experiment/MSAFF-Overfit.yaml → configs/experiment/SIM-Overfit.yaml

@@ -8,7 +8,7 @@ defaults:
 
 logger:
   comet:
-    experiment_name: "MSAFF-Overfit"
+    experiment_name: "SIM-Overfit"
 
 trainer:
   min_epochs: 100

configs/experiment/MSAFFDebug.yaml → configs/experiment/SIMDebug.yaml

@@ -7,7 +7,7 @@ defaults:
 
 logger:
   comet:
-    experiment_name: "MSAFF_COSINE"
+    experiment_name: ""
     disabled: false
 
 

configs/experiment/MSAFF_base_run_FR-2x3GPUs.yaml → configs/experiment/SIM_base_run_FR-2x3GPUs.yaml

@@ -1,10 +1,10 @@
 # @package _global_
 defaults:
-  - MSAFF_base_run_FR.yaml
+  - SIM_base_run_FR.yaml
 
 logger:
   comet:
-    experiment_name: "MSAFF_base_run_FR-2x3GPUs"
+    experiment_name: "SIM_base_run_FR-2x3GPUs"
 
 
 # 2 nodes x 3 GPUs - No gradient accumulation. 

configs/experiment/MSAFF_base_run_FR-MultiGPU.yaml → configs/experiment/SIM_base_run_FR-MultiGPU.yaml

@@ -1,10 +1,10 @@
 # @package _global_
 defaults:
-  - MSAFF_base_run_FR.yaml
+  - SIM_base_run_FR.yaml
 
 logger:
   comet:
-    experiment_name: "MSAFF_base_run_FR-2xGPUs"
+    experiment_name: "SIM_base_run_FR-2xGPUs"
 
 trainer:
   strategy: ddp_find_unused_parameters_false

configs/experiment/MSAFF_base_run_FR.yaml → configs/experiment/SIM_base_run_FR.yaml

@@ -4,7 +4,7 @@ defaults:
 
 logger:
   comet:
-    experiment_name: "MASFF_base_run_FR-(BatchSize10)"
+    experiment_name: "SIM_base_run_FR-(BatchSize10)"
 
 
 datamodule:

configs/model/default.yaml

@@ -21,7 +21,6 @@ monitor: "val_loss"
 false1: 1
 false2: 4
 learning_rate: 0.001
-mode: "feature" # takes "feature" or "feature+decision"
 
 
 defaults:

configs/model/msaff_mlp_model.yaml

@@ -0,0 +1,12 @@
+defaults:
+  - default.yaml
+  - override criterion: BinaryEntropyLoss.yaml
+
+lr: 0.003933709606504788  
+
+neural_net_class_name: "MSNet"
+neural_net_hparams:
+  Inplanes: 32
+
+
+mode: "feature+decision"

configs/model/msaff_model.yaml

@@ -6,3 +6,6 @@ lr: 0.003933709606504788
 neural_net_class_name: "MSNet"
 neural_net_hparams:
   Inplanes: 32
+
+
+mode: "feature"

configs/model/unet_mlp_model.yaml

@@ -0,0 +1,13 @@
+defaults:
+  - default.yaml
+  - override criterion: BinaryEntropyLoss.yaml
+
+
+lr: 0.003933709606504788   # 200-long LR-test-range betw 10^-4 and 3.0
+
+neural_net_class_name: "UNet"
+neural_net_hparams:
+  in_channels: 1
+  init_features: 32
+
+mode: "feature+decision"

configs/model/unet_model.yaml

@@ -5,4 +5,7 @@ lr: 0.003933709606504788   # 200-long LR-test-range betw 10^-4 and 3.0
 
 neural_net_class_name: "UNet"
 neural_net_hparams:
-  Inplanes: 1
+  in_channels: 1
+  init_features: 32
+
+mode: "feature"

configs/model/unetattention_mlp_model.yaml

@@ -0,0 +1,14 @@
+defaults:
+  - default.yaml
+  - override criterion: BinaryEntropyLoss.yaml
+
+
+lr: 0.003933709606504788   # 200-long LR-test-range betw 10^-4 and 3.0
+
+neural_net_class_name: "UNetGatedAttention"
+neural_net_hparams:
+  in_channels: 1
+  init_features: 32
+
+
+mode: "feature+decision"

configs/model/unetattention_model.yaml

@@ -5,4 +5,8 @@ lr: 0.003933709606504788   # 200-long LR-test-range betw 10^-4 and 3.0
 
 neural_net_class_name: "UNetGatedAttention"
 neural_net_hparams:
-  Inplanes: 1
+  in_channels: 1
+  init_features: 32
+
+
+mode: "feature"

docs/source/guides/train_new_model.md

@@ -17,17 +17,17 @@ To test your setup and logging capabilities, you can try overfitting on a single
 To overfit on a single batch for 30 epochs, run:
 
 ```bash
-python run.py experiment=MSAFFDebug
+python run.py experiment=SIMDebug
 ```
 
 ## Training
 
-Define your experiment hyperparameters in an experiment file in the `configs/experiment` folder. You may stem from one of the provided experiment file (e.g. `MSAFF_base_run_FR.yaml`).
+Define your experiment hyperparameters in an experiment file in the `configs/experiment` folder. You may stem from one of the provided experiment file (e.g. `SIM_base_run_FR.yaml`).
 
 To run a typically nominal training,validation and testing, run:
 
 ```bash
-python run.py experiment=MSAFF_base_run_FR
+python run.py experiment=SIM_base_run_FR
 ```
 
 After training, you model best checkpoints and hydra config will be saved in a `DATE/TIME/` subfolder of the `LOGS_DIR` you specified, with an associated hydra `config.yaml`.
@@ -39,7 +39,7 @@ You can perfom this automatically before training by setting `task.auto_lr_find=
 
 ### Multi-GPUs
 
-Multi-GPUs training is supported. Refer to e.g. experiment file `MSAFF_base_run_FR-MultiGPU.yaml` for pytorch lightning flags to activate it. 
+Multi-GPUs training is supported. Refer to e.g. experiment file `SIM_base_run_FR-MultiGPU.yaml` for pytorch lightning flags to activate it. 
 Multi-GPUs training effectively reduces training time by the number of GPUs used. Batch size might need to be reduced to keep a constant number of steps per epoch in DDP.
 
 ## Testing the model
@@ -66,4 +66,4 @@ python run.py experiment=test
 
 ## Inference
 
-To use the checkpointed model to make predictions on new data, refer to section [Performing inference on new data](../tutorials/make_predictions.md).
+To use the checkpointed model to make predictions on new data, refer to section [Performing inference on new data](../tutorials/make_predictions.md).