update tuning

hrdae/conf/experiment/dataloader/basic.yaml

@@ -1,4 +1,4 @@
-batch_size: 96
+batch_size: 64
 train_val_ratio: 0.8
 # # mnist
 # transform_order_train:

hrdae/conf/experiment/model/vr.yaml

@@ -1,7 +1,7 @@
 loss_coef:
   wmse: 1
 phase: "0"
-pred_diff: true
+pred_diff: false
 defaults:
   - /config/experiment/model/vr@_here_
   - [email protected]: wmse

mypy.ini

@@ -15,6 +15,10 @@ ignore_missing_imports = True
 ignore_missing_imports = True
 [mypy-omegaconf.*]
 ignore_missing_imports = True
+[mypy-optuna]
+ignore_missing_imports = True
+[mypy-optuna.*]
+ignore_missing_imports = True
 [mypy-pytorch_tcn]
 ignore_missing_imports = True
 [mypy-pytorch_tcn.*]

requirements.gpu.txt

@@ -1,15 +1,23 @@
+alembic==1.13.1
 antlr4-python3-runtime==4.9.3
+black==24.4.2
+click==8.1.7
+colorlog==6.8.2
 contourpy==1.2.1
 cycler==0.12.1
 filelock==3.14.0
 fonttools==4.53.0
 fsspec==2024.5.0
+greenlet==3.0.3
 hydra-core==1.3.2
+isort==5.13.2
 Jinja2==3.1.4
 kiwisolver==1.4.5
+Mako==1.3.5
 MarkupSafe==2.1.5
 matplotlib==3.9.0
 mpmath==1.3.0
+mypy-extensions==1.0.0
 networkx==3.3
 numpy==1.26.4
 nvidia-cublas-cu12==12.1.3.1
@@ -25,14 +33,20 @@ nvidia-nccl-cu12==2.20.5
 nvidia-nvjitlink-cu12==12.5.40
 nvidia-nvtx-cu12==12.1.105
 omegaconf==2.3.0
+optuna==3.6.1
 packaging==24.0
+pathspec==0.12.1
 pillow==10.3.0
+platformdirs==4.2.2
 pyparsing==3.1.2
 python-dateutil==2.9.0.post0
+pytorch-tcn==1.1.0
 PyYAML==6.0.1
 six==1.16.0
+SQLAlchemy==2.0.30
 sympy==1.12.1
 torch==2.3.0
 torchaudio==2.3.0
 torchvision==0.18.0
+tqdm==4.66.4
 typing_extensions==4.12.1

tuning.py

@@ -1,6 +1,7 @@
 import json
 from dataclasses import asdict
 from pathlib import Path
+from typing import Any
 
 from hrdae.option import TrainExpOption
 from hrdae.dataloaders.transforms import MinMaxNormalizationOption
@@ -35,6 +36,14 @@ def interleave_arrays(
     return result
 
 
+def default(item: Any):
+    match item:
+        case Path():
+            return str(item)
+        case _:
+            raise TypeError(type(item))
+
+
 def objective(trial):
     dataset_option = MovingMNISTDatasetOption(
         root="data",
@@ -63,8 +72,10 @@ def objective(trial):
     phase = trial.suggest_categorical("phase", ["0", "t", "all"])
     network_name = trial.suggest_categorical("network", ["hrdae2d", "rae2d", "rdae2d"])
     if phase == "all":
-        motion_encoder_name = trial.suggest_categorical("motion_encoder", ["conv2d", "normal1d", "rnn1d"])
+        pred_diff = False
+        motion_encoder_name = trial.suggest_categorical("motion_encoder_all", ["conv2d", "normal1d", "rnn1d"])
     else:
+        pred_diff = trial.suggest_categorical("pred_diff", [True, False])
         motion_encoder_name = trial.suggest_categorical("motion_encoder", ["conv2d", "guided1d", "normal1d", "rnn1d", "tsn1d"])
     motion_encoder_num_layers = trial.suggest_int("motion_encoder_num_layers", 0, 3)
     if motion_encoder_name == "rnn1d":
@@ -130,7 +141,7 @@ def objective(trial):
         )
     else:
         raise RuntimeError("unreachable")
-    latent_dim = int(trial.suggest_discrete_uniform("latent_dim", 16, 64, 8))
+    latent_dim = trial.suggest_int("latent_dim", 16, 64, step=8)
     content_encoder_num_layers = trial.suggest_int("content_encoder_encoder_num_layers", 0, 3)
     if network_name == "rae2d":
         network_option = RAE2dOption(
@@ -168,33 +179,33 @@ def objective(trial):
         raise RuntimeError("unreachable")
 
     optimizer_option = AdamOptimizerOption(
-        lr=trial.suggest_loguniform("lr", 1e-5, 1e-2),
+        lr=trial.suggest_float("lr", 1e-5, 1e-2, log=True),
     )
 
     scheduler_option = OneCycleLRSchedulerOption(
-        max_lr=trial.suggest_loguniform("max_lr", 1e-3, 1e-2),
+        max_lr=trial.suggest_float("max_lr", 1e-3, 1e-2, log=True),
     )
 
     model_option = VRModelOption(
         loss_coef={"wmse": 1.0},
         phase=phase,
-        pred_diff=trial.suggest_categorical("pred_diff", [True, False]),
+        pred_diff=pred_diff,
         loss=loss_option,
         network=network_option,
         optimizer=optimizer_option,
         scheduler=scheduler_option,
     )
 
-    result_dir = Path(f"result/tuning/mmnist/{network_name}/{motion_encoder_name}/{trial.number}")
+    result_dir = Path(f"results/tuning/mmnist/{network_name}/{motion_encoder_name}/{trial.number}")
     train_option = TrainExpOption(
         result_dir=result_dir,
         dataloader=dataloader_option,
         model=model_option,
-        n_epoch=50,
+        n_epoch=1,
     )
     result_dir.mkdir(parents=True, exist_ok=True)
     with open(result_dir / "config.json", "w") as f:
-        json.dump(asdict(train_option), f, indent=2)
+        json.dump(asdict(train_option), f, indent=2, default=default)
 
     train_loader, val_loader = create_dataloader(
         dataloader_option,
@@ -210,6 +221,7 @@ def objective(trial):
         val_loader,
         n_epoch=train_option.n_epoch,
         result_dir=result_dir,
+        debug=False,
     )
 
 
@@ -218,7 +230,7 @@ def objective(trial):
 
     study = optuna.create_study(
         study_name="tuning",
-        storage="sqlite:///result/tuning.db",
+        storage="sqlite:///results/tuning.db",
         load_if_exists=True,
     )
     study.optimize(objective, n_trials=100)