[Example] Add TADF材料分子的光电性质预测

TADF/TADF_Est/config/est.yaml

@@ -0,0 +1,82 @@
+defaults: #
+  - ppsci_default #
+  - TRAIN: train_default #
+  - EVAL: eval_default #
+#  - INFER: infer_default #
+  - _self_ #
+
+hydra:
+  run:
+    # dynamic output directory according to running time and override name
+    dir: outputs_Est/${now:%Y-%m-%d}/${now:%H-%M-%S}/${hydra.job.override_dirname} #
+  job:
+    name: ${mode} # name of logfile
+    chdir: false # keep current working directory unchanged
+  callbacks:
+    init_callback: #
+      _target_: ppsci.utils.callbacks.InitCallback #
+  sweep:
+    # output directory for multirun
+    dir: ${hydra.run.dir}
+    subdir: ./
+
+# general settings
+mode: eval # running mode: train/eval #
+seed: 42 #
+output_dir: ${hydra:run.dir} #
+log_freq: 20 #
+use_tbd: false #
+VIV_DATA_PATH: "./f.dat" #
+
+# model settings
+MODEL: #
+#  input_keys: tuple(x.keys()) #
+  output_keys: ["u"] #
+#  num_layers: None #
+#  hidden_size:
+#    - 587
+#    - 256 #
+  activation: "relu" #
+  dropout: 0.5
+
+# training settings
+TRAIN: #
+  epochs: 200 #
+  iters_per_epoch: 20 #
+  save_freq: 100 #
+  eval_during_train: False #
+  batch_size: 8 #
+  learning_rate: 0.0001
+  save_model_path: './est_model.pth'
+  weight_decay: 1e-5
+  pretrained_model_path: null #
+  checkpoint_path: null #
+  k: 9
+  i: 2
+
+# evaluation settings
+EVAL:
+  test_size: 0.1
+  load_model_path: './est_model.pth'
+  seed: 20
+#  pretrained_model_path: null #
+#  batch_size: 32 #
+
+# inference settings
+#INFER: #
+#  pretrained_model_path: "https://github.com/YfB1125/PaddleScience/blob/dev_model/TADF/f/output/checkpoints/latest.pdparams" #
+#  export_path: ./inference/viv #
+#  pdmodel_path: ${INFER.export_path}.pdmodel #
+#  pdiparams_path: ${INFER.export_path}.pdiparams #
+#  input_keys: ${MODEL.input_keys} #
+#  output_keys: ["u"]  #
+#  device: cpu #
+#  engine: native #
+#  precision: fp32 #
+#  ir_optim: true #
+#  min_subgraph_size: 10 #
+#  gpu_mem: 4000 #
+#  gpu_id: 0 #
+#  max_batch_size: 64 #
+#  num_cpu_threads: 4 #
+#  batch_size: 16 #

TADF/TADF_Est/est_ppsci.py

@@ -0,0 +1,177 @@
+import os
+
+import hydra
+import matplotlib.pyplot as plt
+import numpy as np
+import paddle
+import rdkit.Chem as Chem
+from omegaconf import DictConfig
+from rdkit.Chem import AllChem
+from sklearn.decomposition import PCA
+from sklearn.metrics import r2_score
+from sklearn.model_selection import train_test_split
+
+import ppsci
+
+os.environ["HYDRA_FULL_ERROR"] = "1"
+os.environ["KMP_DUPLICATE_LIB_OK"] = "True"
+plt.rcParams["axes.unicode_minus"] = False
+plt.rcParams["font.sans-serif"] = ["DejaVu Sans"]
+
+# 加载数据集
+data = []
+for line in open("./est.dat"):
+    num = float(line.strip())
+    data.append(num)
+smis = []
+for line in open("./smis.txt"):
+    smis.append(line.strip())
+vectors = []
+del_mol = []
+for num in smis:
+    mol = Chem.MolFromSmiles(num)
+    try:
+        fp = AllChem.GetMorganFingerprintAsBitVect(mol, radius=2, nBits=2048)
+        _input = np.array(list(map(float, fp.ToBitString())))
+        vectors.append(_input)
+    except Exception:
+        del_mol.append(num)
+pca = PCA(n_components=0.99)
+pca.fit(vectors)
+Xlist = paddle.to_tensor(pca.transform(vectors))
+
+
+def train(cfg: DictConfig):
+    # 划分数据集
+    def k_fold(k, i, X, Y):
+        fold_size = tuple(X.shape)[0] // k
+        val_start = i * fold_size
+        if i != k - 1:
+            val_end = (i + 1) * fold_size
+            x_val, y_val = X[val_start:val_end], Y[val_start:val_end]
+            x_train = np.concatenate((X[0:val_start], X[val_end:]), axis=0)
+            y_train = np.concatenate((Y[0:val_start], Y[val_end:]), axis=0)
+        else:
+            x_val, y_val = X[val_start:], Y[val_start:]
+            x_train = X[0:val_start]
+            y_train = Y[0:val_start]
+        return x_train, y_train, x_val, y_val
+
+    x_train, y_train, x_test, y_test = k_fold(cfg.TRAIN.k, cfg.TRAIN.i, Xlist, data)
+    # 处理数据集
+    x_train = paddle.to_tensor(x_train, dtype="float32")
+    x = {
+        "key_{}".format(i): paddle.unsqueeze(
+            paddle.to_tensor(x_train[:, i], dtype="float32"), axis=1
+        )
+        for i in range(x_train.shape[1])
+    }
+    y_train = paddle.unsqueeze(paddle.to_tensor(y_train, dtype="float32"), axis=1)
+
+    # 构建约束
+    bc_sup = ppsci.constraint.SupervisedConstraint(
+        dataloader_cfg={
+            "dataset": {
+                "input": x,
+                "label": {"u": y_train},
+                "name": "IterableNamedArrayDataset",
+            },
+            "batch_size": cfg.TRAIN.batch_size,
+        },
+        loss=ppsci.loss.MSELoss("mean"),
+        name="bc_sup",
+    )
+
+    # 设置模型
+    hidden_size = [587, 256]
+    num_layers = None
+    # 实例化模型
+    model = ppsci.arch.DNN(
+        input_keys=tuple(x.keys()),
+        hidden_size=hidden_size,
+        num_layers=num_layers,
+        **cfg.MODEL,
+    )
+    optimizer = ppsci.optimizer.optimizer.Adam(
+        cfg.TRAIN.learning_rate,
+        beta1=(0.9, 0.99)[0],
+        beta2=(0.9, 0.99)[1],
+        weight_decay=cfg.TRAIN.weight_decay,
+    )(model)
+    # 构建Solver
+    solver = ppsci.solver.Solver(
+        model,
+        constraint={
+            "bc_sup": bc_sup,
+        },
+        optimizer=optimizer,
+        epochs=cfg.TRAIN.epochs,
+        eval_during_train=False,
+        iters_per_epoch=cfg.TRAIN.iters_per_epoch,
+        seed=cfg.seed,
+    )
+    try:
+        solver.train()
+    except Exception as ex:
+        print("error", ex)
+    paddle.save(model.state_dict(), cfg.TRAIN.save_model_path)
+
+
+# 进行测试
+def eval(cfg: DictConfig):
+    # 重新划分数据集
+    x_train, x_test, y_train, y_test = train_test_split(
+        Xlist, data, test_size=cfg.EVAL.test_size, random_state=cfg.EVAL.seed
+    )
+    x = {
+        "key_{}".format(i): paddle.unsqueeze(
+            paddle.to_tensor(x_test[:, i], "float32"), axis=1
+        )
+        for i in range(x_test.shape[1])
+    }
+    hidden_size = [587, 256]
+    num_layers = None
+    model = ppsci.arch.DNN(
+        input_keys=tuple(x.keys()),
+        hidden_size=hidden_size,
+        num_layers=num_layers,
+        **cfg.MODEL,
+    )
+    model.set_state_dict(paddle.load(cfg.EVAL.load_model_path))
+    ytest = paddle.unsqueeze(paddle.to_tensor(y_test, dtype="float32"), axis=1)
+    ypred = model(x)
+    ytest = {"u": ytest}
+
+    # 计算损失
+    loss = ppsci.metric.MAE()
+    MAE = loss(ypred, ytest).get("u").numpy()
+    loss = ppsci.metric.RMSE()
+    RMSE = loss(ypred, ytest).get("u").numpy()
+    ypred = ypred.get("u").numpy()
+    ytest = ytest.get("u").numpy()
+    R2 = r2_score(ytest, ypred)
+    print("MAE", MAE)
+    print("RMSE", RMSE)
+    print("R2", R2)
+
+    # 可视化
+    plt.scatter(ytest, ypred, s=15, color="royalblue", marker="s", linewidth=1)
+    plt.plot([ytest.min(), ytest.max()], [ytest.min(), ytest.max()], "r-", lw=1)
+    plt.legend(title="R²={:.3f}\n\nMAE={:.3f}".format(R2, MAE))
+    plt.xlabel("Test ΔEst(eV)")
+    plt.ylabel("Predicted ΔEst(eV)")
+    plt.show()
+
+
+@hydra.main(version_base=None, config_path="./config", config_name="est.yaml")
+def main(cfg: DictConfig):
+    if cfg.mode == "train":
+        train(cfg)
+    elif cfg.mode == "eval":
+        eval(cfg)
+    else:
+        raise ValueError(f"cfg.mode should in ['train', 'eval'], but got '{cfg.mode}'")
+
+
+if __name__ == "__main__":
+    main()

TADF/TADF_Est/outputs_Est/2024-10-15/23-53-31/.hydra/config.yaml

@@ -0,0 +1,47 @@
+mode: train
+output_dir: ${hydra:run.dir}
+log_freq: 20
+seed: 42
+use_vdl: false
+use_tbd: false
+wandb_config: null
+use_wandb: false
+device: gpu
+use_amp: false
+amp_level: O1
+to_static: false
+prim: false
+log_level: info
+TRAIN:
+  epochs: 200
+  iters_per_epoch: 20
+  update_freq: 1
+  save_freq: 100
+  eval_during_train: false
+  start_eval_epoch: 1
+  eval_freq: 1
+  checkpoint_path: null
+  pretrained_model_path: null
+  ema: null
+  swa: null
+  batch_size: 8
+  learning_rate: 0.0001
+  save_model_path: ./est_model.pth
+  weight_decay: 1.0e-05
+  k: 9
+  i: 2
+EVAL:
+  pretrained_model_path: null
+  eval_with_no_grad: false
+  compute_metric_by_batch: false
+  batch_size: 256
+  test_size: 0.1
+  load_model_path: ./est_model.pth
+  seed: 20
+INFER: null
+VIV_DATA_PATH: ./f.dat
+MODEL:
+  output_keys:
+  - u
+  activation: relu
+  dropout: 0.5