【Hackathon 8th No.16】 data_efficient_nopt 论文复现

README.md

@@ -48,6 +48,7 @@ PaddleScience 是一个基于深度学习框架 PaddlePaddle 开发的科学计
 | 微分方程 | [若斯叻方程](https://paddlescience-docs.readthedocs.io/zh-cn/latest/zh/examples/rossler) | 数据驱动 | Transformer-Physx | 监督学习 | [Data](https://github.com/zabaras/transformer-physx) | [Paper](https://arxiv.org/abs/2010.03957) |
 | 算子学习 | [DeepONet](https://paddlescience-docs.readthedocs.io/zh-cn/latest/zh/examples/deeponet) | 数据驱动 | MLP | 监督学习 | [Data](https://deepxde.readthedocs.io/en/latest/demos/operator/antiderivative_unaligned.html) | [Paper](https://export.arxiv.org/pdf/1910.03193.pdf) |
 | 微分方程 | [梯度增强的物理知识融合 PDE 求解](https://github.com/PaddlePaddle/PaddleScience/blob/develop/examples/gpinn/poisson_1d.py) | 机理驱动 | gPINN | 无监督学习 | - |  [Paper](https://doi.org/10.1016/j.cma.2022.114823) |
+| 微分方程 | [PDE 求解](https://paddlescience-docs.readthedocs.io/zh-cn/latest/zh/examples/data_efficient_nopt) | 数据驱动 | FNO/Transformer | 无监督学习 | - |  [Paper](https://arxiv.org/abs/2402.15734) |
 | 积分方程 | [沃尔泰拉积分方程](https://paddlescience-docs.readthedocs.io/zh-cn/latest/zh/examples/volterra_ide) | 机理驱动 | MLP | 无监督学习 | - | [Project](https://github.com/lululxvi/deepxde/blob/master/examples/pinn_forward/Volterra_IDE.py) |
 | 微分方程 | [分数阶微分方程](https://github.com/PaddlePaddle/PaddleScience/blob/develop/examples/fpde/fractional_poisson_2d.py) | 机理驱动 | MLP | 无监督学习 | - | - |
 | 光孤子 | [Optical soliton](https://paddlescience-docs.readthedocs.io/zh-cn/latest/zh/examples/nlsmb) | 机理驱动 | MLP | 无监督学习 | - | [Paper](https://doi.org/10.1007/s11071-023-08824-w)|

docs/index.md

@@ -83,6 +83,7 @@
 | 微分方程 | [若斯叻方程](./zh/examples/rossler.md) | 数据驱动 | Transformer-Physx | 监督学习 | [Data](https://github.com/zabaras/transformer-physx) | [Paper](https://arxiv.org/abs/2010.03957) |
 | 算子学习 | [DeepONet](./zh/examples/deeponet.md) | 数据驱动 | MLP | 监督学习 | [Data](https://deepxde.readthedocs.io/en/latest/demos/operator/antiderivative_unaligned.html) | [Paper](https://export.arxiv.org/pdf/1910.03193.pdf) |
 | 微分方程 | [梯度增强的物理知识融合 PDE 求解](https://github.com/PaddlePaddle/PaddleScience/blob/develop/examples/gpinn/poisson_1d.py) | 机理驱动 | gPINN | 无监督学习 | - |  [Paper](https://doi.org/10.1016/j.cma.2022.114823) |
+| 微分方程 | [PDE 求解](./zh/examples/data_efficient_nopt.md) | 数据驱动 | FNO/Transformer | 无监督学习 | - |  [Paper](https://arxiv.org/abs/2402.15734) |
 | 积分方程 | [沃尔泰拉积分方程](./zh/examples/volterra_ide.md) | 机理驱动 | MLP | 无监督学习 | - | [Project](https://github.com/lululxvi/deepxde/blob/master/examples/pinn_forward/Volterra_IDE.py) |
 | 微分方程 | [分数阶微分方程](https://github.com/PaddlePaddle/PaddleScience/blob/develop/examples/fpde/fractional_poisson_2d.py) | 机理驱动 | MLP | 无监督学习 | - | - |
 | 光孤子 | [Optical soliton](./zh/examples/nlsmb.md) | 机理驱动 | MLP | 无监督学习 | - | [Paper](https://doi.org/10.1007/s11071-023-08824-w)|

examples/data_efficient_nopt/config/data_efficient_nopt_fno_poisson.yaml

@@ -0,0 +1,325 @@
+# general settings
+mode: train
+seed: 42
+
+# training settings
+run_name: r0
+use_ddp: False
+config: pois-64-pretrain-e1_20_m3
+sweep_id: ''
+logdir: exp
+output_dir: ${hydra:run.dir}
+
+train_config:
+  default: &DEFAULT
+    num_data_workers: 4
+    # model
+    model: 'fno'
+    depth: 5
+    in_dim: 2
+    out_dim: 1
+    dropout: 0
+    # data/domain
+    Lx: !!float 1.0
+    Ly: !!float 1.0
+    nx: 256
+    ny: 256
+    # optimization
+    loss_style: 'mean'
+    loss_func: 'mse'
+    optimizer: 'adam'
+    scheduler: 'none'
+    learning_rate: !!float 1.0
+    max_epochs: 500
+    scheduler_epochs: 500
+    weight_decay: 0
+    batch_size: 25
+    # misc
+    log_to_screen: !!bool True
+    save_checkpoint: !!bool False
+    seed: 0
+    plot_figs: !!bool False
+    pack_data: !!bool False
+    # Weights & Biases
+    entity: 'entity_name'
+    project: 'proj_name'
+    group: 'poisson'
+    log_to_wandb: !!bool False
+    distill: !!bool False
+    subsample: 1
+    exp_dir: './exp/'
+    tie_fields: !!bool False # Whether to use 1 embedding per field per data
+    use_all_fields: !!bool True # Prepopulate the field metadata dictionary from dictionary in datasets
+    tie_batches: !!bool False # Force everything in batch to come from one dset
+    model_type: fno
+    pretrained: False
+    warmup_steps: 0
+    epoch_size: 1
+    accum_grad: 1
+    enable_amp: !!bool False
+    log_interval: 1
+    checkpoint_save_interval: 1000
+    debug_grad: False
+
+  poisson: &poisson
+    <<: *DEFAULT
+    n_demos: 0
+    batch_size: 512
+    nx: 128
+    ny: 128
+    save_checkpoint: !!bool True
+    max_epochs: 500
+    scheduler: 'cosine'
+
+    model: 'fno'
+    layers: [64, 64, 64, 64, 64]
+    modes1: [65, 65, 65, 65]
+    modes2: [65, 65, 65, 65]
+    fc_dim: 256
+
+    in_dim: 4
+    out_dim: 1
+    mode_cut: 16
+    embed_cut: 64
+    fc_cut: 2
+
+    optimizer: 'adam'
+
+    learning_rate: 1E-3
+    pack_data: !!bool False
+
+
+  poisson-64-scale-e5_15: &poisson_64_e5_15
+    <<: *poisson
+    train_path:  'data/possion_64/poisson_64_e5_15_train.h5'
+    val_path:    'data/possion_64/poisson_64_e5_15_val.h5'
+    test_path:   'data/possion_64/poisson_64_e5_15_test.h5'
+    scales_path: 'data/possion_64/poisson_64_e5_15_train_scale.npy'
+    train_rand_idx_path: 'data/possion_64/train_rand_idx.npy'
+    batch_size: 128
+    log_to_wandb: !!bool False
+    learning_rate: 1E-3
+
+    mode_cut: 32
+    embed_cut: 64
+    fc_cut: 2
+    subsample: 1
+    nx: 64
+    ny: 64
+
+    pt: "train"
+    pt_split: [46080, 8192]
+    pretrained: False
+
+
+  pois-64-pretrain-e1_20: &pois_64_e1_20_pt
+    <<: *poisson
+    train_path:   'data/possion_64/poisson_64_e1_20_train.h5'
+    val_path:     'data/possion_64/poisson_64_e1_20_val.h5'
+    test_path:    'data/possion_64/poisson_64_e1_20_test.h5'
+    scales_path:  'data/possion_64/poisson_64_e1_20_train_scale.npy'
+    train_rand_idx_path: 'data/possion_64/train_rand_idx.npy'
+    batch_size: 128
+    log_to_wandb: !!bool False
+    mode_cut: 32
+    embed_cut: 64
+    fc_cut: 2
+    subsample: 1
+    nx: 64
+    ny: 64
+    learning_rate: 1E-3
+    pt: "pretrain"
+    pt_split: [46080, 8192]
+    blur: [0, 1]
+
+
+  pois_64_finetune_e5_15: &pois_64_e5_15_ft
+    <<: *poisson
+    train_path:   'data/possion_64/poisson_64_e5_15_train.h5'
+    val_path:     'data/possion_64/poisson_64_e5_15_val.h5'
+    test_path:    'data/possion_64/poisson_64_e5_15_test.h5'
+    scales_path:  'data/possion_64/poisson_64_e5_15_train_scale.npy'
+    train_rand_idx_path: 'data/possion_64/train_rand_idx.npy'
+    batch_size: 128
+    log_to_wandb: !!bool False
+    mode_cut: 32
+    embed_cut: 64
+    fc_cut: 2
+    subsample: 1
+    nx: 64
+    ny: 64
+    learning_rate: 1E-3
+    pt: "train"
+    pt_split: [46080, 8192]
+    fix_backbone: False
+    resuming: False
+    pretrained: True
+    pretrained_ckpt_path: /pretrained_ckpt_path/training_checkpoints/ckpt.tar
+
+  pois-64-e5_15_ft0: &pois_64_e5_15_ft0
+    <<: *pois_64_e5_15_ft
+    subsample: 1
+
+  pois-64-e5_15_ft1: &pois_64_e5_15_ft1
+    <<: *pois_64_e5_15_ft
+    subsample: 2
+
+  pois-64-e5_15_ft2: &pois_64_e5_15_ft2
+    <<: *pois_64_e5_15_ft
+    subsample: 4
+
+  pois-64-e5_15_ft3: &pois_64_e5_15_ft3
+    <<: *pois_64_e5_15_ft
+    subsample: 8
+
+  pois-64-e5_15_ft4: &pois_64_e5_15_ft4
+    <<: *pois_64_e5_15_ft
+    subsample: 16
+
+  pois-64-e5_15_ft5: &pois_64_e5_15_ft5
+    <<: *pois_64_e5_15_ft
+    subsample: 32
+
+  pois-64-e5_15_ft6: &pois_64_e5_15_ft6
+    <<: *pois_64_e5_15_ft
+    subsample: 64
+
+  pois-64-e5_15_ft7: &pois_64_e5_15_ft7
+    <<: *pois_64_e5_15_ft
+    subsample: 128
+    batch_size: 64
+
+  pois-64-e5_15_ft8: &pois_64_e5_15_ft8
+    <<: *pois_64_e5_15_ft
+    subsample: 256
+    batch_size: 32
+
+  pois-64-e5_15_ft9: &pois_64_e5_15_ft9
+    <<: *pois_64_e5_15_ft
+    subsample: 512
+    batch_size: 16
+
+  pois-64-pretrain-e1_20_m0: &pois-64-e1_20_pt_m0
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.
+
+  pois-64-pretrain-e1_20_m1: &pois-64-e1_20_pt_m1
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.1
+
+  pois-64-pretrain-e1_20_m2: &pois-64-e1_20_pt_m2
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.2
+
+  pois-64-pretrain-e1_20_m3: &pois-64-e1_20_pt_m3
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.3
+
+  pois-64-pretrain-e1_20_m4: &pois-64-e1_20_pt_m4
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.4
+
+  pois-64-pretrain-e1_20_m5: &pois-64-e1_20_pt_m5
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.5
+
+  pois-64-pretrain-e1_20_m6: &pois-64-e1_20_pt_m6
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.6
+
+  pois-64-pretrain-e1_20_m7: &pois-64-e1_20_pt_m7
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.7
+
+  pois-64-pretrain-e1_20_m8: &pois-64-e1_20_pt_m8
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.8
+
+  pois-64-pretrain-e1_20_m9: &pois-64-e1_20_pt_m9
+    <<: *pois_64_e1_20_pt
+    mask_ratio: 0.9
+
+
+
+  poisson-64-e5_15_bsln: &pois_64_e5_15_baseline
+    <<: *poisson_64_e5_15
+
+  # 8192
+  poisson-64-e5_15_b0: &pois_64_e5_15_ss4
+    <<: *pois_64_e5_15_baseline
+    subsample: 1
+
+  poisson-64-e5_15_b1: &pois_64_e5_15_ss8
+    <<: *pois_64_e5_15_baseline
+    subsample: 2
+
+  poisson-64-e5_15_b2: &pois_64_e5_15_ss16
+    <<: *pois_64_e5_15_baseline
+    subsample: 4
+
+  poisson-64-e5_15_b3: &pois_64_e5_15_ss32
+    <<: *pois_64_e5_15_baseline
+    subsample: 8
+
+  poisson-64-e5_15_b4: &pois_64_e5_15_ss64
+    <<: *pois_64_e5_15_baseline
+    subsample: 16
+
+  poisson-64-e5_15_b5: &pois_64_e5_15_ss128
+    <<: *pois_64_e5_15_baseline
+    subsample: 32
+
+  poisson-64-e5_15_b6: &pois_64_e5_15_ss256
+    <<: *pois_64_e5_15_baseline
+    subsample: 64
+
+  poisson-64-e5_15_b7: &pois_64_e5_15_ss512
+    <<: *pois_64_e5_15_baseline
+    subsample: 128
+    batch_size: 64
+
+
+# inference settings
+ckpt_path: data/pd_finetune_b01_m0_n8192.tar
+num_demos: 1
+tqdm: False
+save_pred: False
+
+infer_config:
+  train_path: 'data/possion_64/poisson_64_e15_50_train.h5' # pick demos
+  test_path: 'data/possion_64/poisson_64_e15_50_test.h5'
+  scales_path: 'data/possion_64/poisson_64_e5_15_train_scale.npy'
+  ckpt_path: data/possion_64/finetune_b01_m0_n8192.pdparams
+
+  num_data_workers: 4
+  subsample: 1
+  num_demos: 0
+  shuffle: False
+  nx: 64
+  nt: 64
+  Lx: !!float 1.0
+  Ly: !!float 1.0
+  pack_data: !!bool False
+
+  model: 'fno'
+  layers: [64, 64, 64, 64, 64]
+  modes1: [65, 65, 65, 65]
+  modes2: [65, 65, 65, 65]
+  fc_dim: 128
+
+  in_dim: 4
+  out_dim: 1
+  mode_cut: 32
+  embed_cut: 64
+  fc_cut: 2
+  dropout: 0
+
+  fix_backbone: True
+
+  loss_func: mse
+
+  batch_size: 1
+  loss_style: sum
+
+  log_to_wandb: !!bool False
+  logdir: ./log