update code

Author: HydrogenSulfate

examples/NLS-MB/NLS-MB_optical_rogue_wave.py

@@ -258,7 +258,6 @@ def train(cfg: DictConfig):
         epochs=EPOCHS,
         equation=equation,
         validator=validator,
-        cfg=cfg,
     )
     # train model
     solver.train()

examples/ldc/conf/ldc2d_modulus.yaml

@@ -18,13 +18,15 @@
 
 # This file is for step1: training a embedding model.
 # This file is based on PaddleScience/ppsci API.
+from os import path as osp
 
 import hydra
 import numpy as np
 import paddle
 from omegaconf import DictConfig
 
 import ppsci
+from ppsci.utils import logger
 
 
 def get_mean_std(data: np.ndarray):
@@ -38,6 +40,11 @@ def get_mean_std(data: np.ndarray):
 
 
 def train(cfg: DictConfig):
+    # set random seed for reproducibility
+    ppsci.utils.misc.set_random_seed(cfg.seed)
+    # initialize logger
+    logger.init_logger("ppsci", osp.join(cfg.output_dir, f"{cfg.mode}.log"), "info")
+
     weights = (1.0 * (cfg.TRAIN_BLOCK_SIZE - 1), 1.0e4 * cfg.TRAIN_BLOCK_SIZE)
     regularization_key = "k_matrix"
     # manually build constraint(s)
@@ -130,9 +137,12 @@ def train(cfg: DictConfig):
     solver = ppsci.solver.Solver(
         model,
         constraint,
-        optimizer=optimizer,
+        cfg.output_dir,
+        optimizer,
+        epochs=cfg.TRAIN.epochs,
+        iters_per_epoch=ITERS_PER_EPOCH,
+        eval_during_train=True,
         validator=validator,
-        cfg=cfg,
     )
     # train model
     solver.train()
@@ -141,6 +151,11 @@ def train(cfg: DictConfig):
 
 
 def evaluate(cfg: DictConfig):
+    # set random seed for reproducibility
+    ppsci.utils.misc.set_random_seed(cfg.seed)
+    # initialize logger
+    logger.init_logger("ppsci", osp.join(cfg.output_dir, f"{cfg.mode}.log"), "info")
+
     weights = (1.0 * (cfg.TRAIN_BLOCK_SIZE - 1), 1.0e4 * cfg.TRAIN_BLOCK_SIZE)
     regularization_key = "k_matrix"
     # manually build constraint(s)

examples/ldc/conf/ldc2d_modulus_importance_sampling.yaml

@@ -18,6 +18,7 @@
 
 # This file is for step2: training a transformer model, based on frozen pretrained embedding model.
 # This file is based on PaddleScience/ppsci API.
+from os import path as osp
 from typing import Dict
 
 import hydra
@@ -26,6 +27,7 @@
 
 import ppsci
 from ppsci.arch import base
+from ppsci.utils import logger
 from ppsci.utils import save_load
 
 
@@ -55,6 +57,9 @@ def train(cfg: DictConfig):
     # valid time-series: 64      time-steps: 1024   block-size: 256 stride: 1024
     # test  time-series: 256     time-steps: 1024
     # set random seed for reproducibility
+    ppsci.utils.misc.set_random_seed(cfg.seed)
+    # initialize logger
+    logger.init_logger("ppsci", osp.join(cfg.output_dir, f"{cfg.mode}.log"), "info")
 
     embedding_model = build_embedding_model(cfg.EMBEDDING_MODEL_PATH)
     output_transform = OutputTransform(embedding_model)
@@ -147,10 +152,14 @@ def train(cfg: DictConfig):
     solver = ppsci.solver.Solver(
         model,
         constraint,
-        optimizer=optimizer,
+        cfg.output_dir,
+        optimizer,
+        epochs=cfg.TRAIN.epochs,
+        iters_per_epoch=ITERS_PER_EPOCH,
+        eval_during_train=cfg.TRAIN.eval_during_train,
+        eval_freq=cfg.TRAIN.eval_freq,
         validator=validator,
         visualizer=visualizer,
-        cfg=cfg,
     )
     # train model
     solver.train()
@@ -161,6 +170,9 @@ def train(cfg: DictConfig):
 
 
 def evaluate(cfg: DictConfig):
+    # directly evaluate pretrained model(optional)
+    logger.init_logger("ppsci", osp.join(cfg.output_dir, f"{cfg.mode}.log"), "info")
+
     embedding_model = build_embedding_model(cfg.EMBEDDING_MODEL_PATH)
     output_transform = OutputTransform(embedding_model)
 

examples/lorenz/train_enn.py

@@ -29,40 +29,40 @@ hydra:
   sweep:
     # output directory for multirun
     dir: ${hydra.run.dir}
-seed: 1234
+
+mode: train
 output_dir: ${hydra:run.dir}
+seed: 1234
 data_dir: ./data/
-log_freq: 20
+log_freq: 5000
+ntrain: 11333
+nb_train: 2952
+n0_train: 986
+alpha: 100
+beta: 100
+re: 999.35
+
 MODEL:
   input_keys: ["x", "y","z","t"]
   output_keys: ["u", "v", "w","p"]
   num_layers: 10
   hidden_size: 300
   activation: "tanh"
   weight_norm: True
-mode: train
-ntrain: 11333
-nb_train: 2952
-n0_train: 986
-alpha: 100
-beta: 100
-re: 999.35
-epochs: 15250
 TRAIN:
-  log_freq: 5000
+  epochs: 15250
   eval_freq: 5000
   save_freq: 5000
-  eval_with_no_grad: true
+  iters_per_epoch: 150
   lr_scheduler:
-    epochs: 15250
+    epochs: ${TRAIN.epochs}
     decay_epochs: [250, 4500, 5000, 5500]
-    iters_per_epoch: 150
     values: [1e-3, 1e-4, 1e-5, 1e-6, 1e-7]
+
 EVAL:
   pretrained_model_path: null
   eval_with_no_grad: true
 
-
 INFER:
   pretrained_model_path: https://paddle-org.bj.bcebos.com/paddlescience/models/nsfnet/nsfnet4.pdparams
   export_path: ./inference/VP_NSFNet4

examples/lorenz/train_transformer.py

@@ -140,6 +140,7 @@ def output_trans_p(input, out):
     solver = ppsci.solver.Solver(
         model,
         constraint,
+        iters_per_epoch=ITERS_PER_EPOCH,
         optimizer=optimizer,
         equation=equation,
         cfg=cfg,

examples/nsfnet/conf/VP_NSFNet4.yaml

@@ -38,9 +38,10 @@ output_dir: ${hydra:run.dir}
 log_freq: 20
 DATASET_PATH: ./datasets/tempoGAN/2d_train.mat
 DATASET_PATH_VALID: ./datasets/tempoGAN/2d_valid.mat
+use_amp: true
+amp_level: O2
 
 # set working condition
-USE_AMP: true
 USE_SPATIALDISC: true
 USE_TEMPODISC: true
 WEIGHT_GEN: [5.0, 0.0, 1.0]  # lambda_l1, lambda_l2, lambda_t
@@ -96,7 +97,6 @@ TRAIN:
     gamma: 0.05
     by_epoch: true
   eval_during_train: false
-  amp_level: O2
   pretrained_model_path: null
   checkpoint_path: null
 

examples/pipe/poiseuille_flow.py

@@ -223,21 +223,21 @@ def train(cfg: DictConfig):
     solver_gen = ppsci.solver.Solver(
         model_list,
         constraint_gen,
-        optimizer_gen,
+        optimizer=optimizer_gen,
         cfg=cfg,
     )
     if cfg.USE_SPATIALDISC:
         solver_disc = ppsci.solver.Solver(
             model_list,
             constraint_disc,
-            optimizer_disc,
+            optimizer=optimizer_disc,
             cfg=cfg,
         )
     if cfg.USE_TEMPODISC:
         solver_disc_tempo = ppsci.solver.Solver(
             model_list,
             constraint_disc_tempo,
-            optimizer_disc_tempo,
+            optimizer=optimizer_disc_tempo,
             cfg=cfg,
         )