gpbench 1.0.0__py3-none-any.whl

gpbench/method_reg/EIR/EIR.py

@@ -0,0 +1,258 @@
+import time
+import argparse
+import pynvml
+import psutil
+import os
+import torch
+import random
+import swanlab
+import numpy as np
+import pandas as pd
+from torch.utils.data import DataLoader,TensorDataset
+from sklearn.model_selection import KFold, train_test_split
+import  sys
+sys.path.append("..")
+from .utils.models_locally_connected import LCLModel
+from .utils.common import DataDimensions
+from sklearn.metrics import  mean_squared_error, mean_absolute_error, r2_score
+from scipy.stats import pearsonr
+from .EIR_Hyperparameters import Hyperparameter
+os.environ['CUDA_VISIBLE_DEVICE'] = '0'
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Argument parser")
+    parser.add_argument('--methods', type=str, default='EIR/', help='Random seed')
+    parser.add_argument('--species', type=str, default='', help='Species name')
+    parser.add_argument('--phe', type=str, default='', help='Dataset name')
+    parser.add_argument('--data_dir', type=str, default='../../data/')
+    parser.add_argument('--result_dir', type=str, default='result/')
+
+    parser.add_argument('--epochs', type=int, default=1000, help='Number of training rounds')
+    parser.add_argument('--batch_size', type=int, default=32, help='Batch size')
+    parser.add_argument('--learning_rate', type=float, default=0.01, help='Learning rate')
+    parser.add_argument('--patience', type=int, default=50, help='Patience for early stopping')
+    args = parser.parse_args()
+    return args
+
+
+def load_data(args):
+    xData = np.load(os.path.join(args.data_dir, args.species, 'genotype.npz'))["arr_0"]
+    yData = np.load(os.path.join(args.data_dir, args.species, 'phenotype.npz'))["arr_0"]
+    names = np.load(os.path.join(args.data_dir, args.species, 'phenotype.npz'))["arr_1"]
+
+    nsample = xData.shape[0]
+    nsnp = xData.shape[1]
+    print("Number of samples: ", nsample)
+    print("Number of SNPs: ", nsnp)
+    return xData, yData, nsample, nsnp, names
+
+def get_gpu_mem_by_pid(pid):
+    procs = pynvml.nvmlDeviceGetComputeRunningProcesses(handle)
+    for p in procs:
+        if p.pid == pid:
+            return p.usedGpuMemory / 1024**2
+    return 0.0
+
+def init():
+    seed = 42
+    os.environ['PYTHONHASHSEED'] = str(seed)
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed(seed)
+
+def set_seed(seed=42):
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+
+def one_hot_encode_ATCG(char):
+    c = char
+    if char==0:
+        return [1,0,0,0]
+    elif char==1:
+        return [0,1,0,0]
+    elif char==2:
+        return [0,0,1,0]
+    else:
+        return [0,0,0,1]
+
+def one_hot_seq(df:pd.DataFrame, nsnp:int):
+    one_hot_df = df.applymap(lambda x:one_hot_encode_ATCG(x))
+    one_hot_df = one_hot_df.values.tolist()
+    one_hot_df = np.array(one_hot_df)
+    one_hot_df = np.reshape(one_hot_df, (one_hot_df.shape[0],-1))
+    tensor_data = torch.Tensor(one_hot_df)
+    return tensor_data
+
+
+def train_model(model, train_loader, valid_loader, optimizer, criterion, num_epochs, patience, device):
+
+    model.to(device)
+    best_loss = float('inf')
+    best_state = None
+    trigger_times = 0
+
+    for epoch in range(num_epochs):
+        model.train()
+        train_loss = 0.0
+        for inputs, labels in train_loader:
+            inputs, labels = inputs.to(device), labels.to(device)
+            optimizer.zero_grad()
+            outputs = model(inputs)
+            labels = labels.unsqueeze(1)
+            loss = criterion(outputs, labels)
+            loss.backward()
+            optimizer.step()
+            train_loss += loss.item() * inputs.size(0)
+
+        model.eval()
+        valid_loss = 0.0
+        with torch.no_grad():
+            for inputs, labels in valid_loader:
+                inputs, labels = inputs.to(device), labels.to(device)
+                outputs = model(inputs)
+                labels = labels.unsqueeze(1)
+                loss = criterion(outputs, labels)
+                valid_loss += loss.item() * inputs.size(0)
+
+        train_loss /= len(train_loader.dataset)
+        valid_loss /= len(valid_loader.dataset)
+
+        # ---------- Early stopping ----------
+        if valid_loss < best_loss:
+            best_loss = valid_loss
+            best_state = model.state_dict()
+            trigger_times = 0
+        else:
+            trigger_times += 1
+            if trigger_times >= patience:
+                print(f"Early stopping at epoch {epoch+1}")
+                break
+
+    if best_state is not None:
+        model.load_state_dict(best_state)
+    return best_loss
+    
+def predict(model, test_loader, device):
+    model.eval()
+    y_pred = []
+    with torch.no_grad():
+        for inputs, _ in test_loader:
+            inputs = inputs.to(device)
+            outputs = model(inputs)  
+            y_pred.append(outputs.cpu().numpy())
+    y_pred = np.concatenate(y_pred, axis=0)
+    y_pred = np.squeeze(y_pred)
+    return y_pred
+
+
+def run_nested_cv(args, data, label, nsnp, device):
+    result_dir = os.path.join(args.result_dir, args.methods + args.species + args.phe)
+    os.makedirs(result_dir, exist_ok=True)
+    print("Starting 10-fold cross-validation...")
+    kf = KFold(n_splits=10, shuffle=True, random_state=42)
+    data = data.reshape(data.shape[0], 1, nsnp, -1)
+
+    all_mse, all_mae, all_r2, all_pcc = [], [], [], []
+    time_star = time.time()
+    for fold, (train_index, test_index) in enumerate(kf.split(data)):
+        print(f"Running fold {fold}...")
+        process = psutil.Process(os.getpid())
+        fold_start_time = time.time()
+
+        X_train, X_test = data[train_index], data[test_index]
+        y_train, y_test = label[train_index], label[test_index]
+
+        X_train_sub, X_valid, y_train_sub, y_valid = train_test_split(X_train, y_train, test_size=0.1, random_state=42)
+        
+        x_train_tensor = torch.from_numpy(X_train_sub).float().to(device)
+        y_train_tensor = torch.from_numpy(y_train_sub).float().to(device)
+        x_valid_tensor = torch.from_numpy(X_valid).float().to(device)
+        y_valid_tensor = torch.from_numpy(y_valid).float().to(device)
+        x_test_tensor = torch.from_numpy(X_test).float().to(device)
+        y_test_tensor = torch.from_numpy(y_test).float().to(device)
+
+        train_data = TensorDataset(x_train_tensor, y_train_tensor)
+        valid_data = TensorDataset(x_valid_tensor, y_valid_tensor)
+        test_data = TensorDataset(x_test_tensor, y_test_tensor)
+
+        train_loader = DataLoader(train_data, args.batch_size, shuffle=True)
+        valid_loader = DataLoader(valid_data, args.batch_size, shuffle=False)
+        test_loader = DataLoader(test_data, args.batch_size, shuffle=False)
+
+        model = LCLModel(DataDimensions(channels=1, height=nsnp, width=1)).to(device)
+        optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate, weight_decay=1e-4)
+        loss_fn = torch.nn.MSELoss()
+
+        train_model(model, train_loader, valid_loader, optimizer, loss_fn, args.epochs, args.patience, device)
+        y_pred = predict(model, test_loader, device)
+
+        mse = mean_squared_error(y_test, y_pred)
+        r2 = r2_score(y_test, y_pred)
+        mae = mean_absolute_error(y_test, y_pred)
+        pcc, _ = pearsonr(y_test, y_pred)
+
+        all_mse.append(mse)
+        all_r2.append(r2)
+        all_mae.append(mae)
+        all_pcc.append(pcc)
+
+        fold_time = time.time() - fold_start_time
+        fold_gpu_mem =  get_gpu_mem_by_pid(os.getpid())
+        fold_cpu_mem = process.memory_info().rss / 1024**2
+        print(f'Fold {fold}:  Corr={pcc:.4f}, MAE={mae:.4f}, MSE={mse:.4f}, R2={r2:.4f}, Time={fold_time:.2f}s, '
+              f'GPU={fold_gpu_mem:.2f}MB, CPU={fold_cpu_mem:.2f}MB')
+
+        torch.cuda.empty_cache()
+        torch.cuda.reset_peak_memory_stats()
+        results_df = pd.DataFrame({'Y_test': y_test, 'Y_pred': y_pred})
+        results_df.to_csv(os.path.join(result_dir, f"fold{fold}.csv"), index=False)
+
+    print("\n===== Cross-validation summary =====")
+    print(f"Average PCC: {np.mean(all_pcc):.4f} ± {np.std(all_pcc):.4f}")
+    print(f"Average MAE: {np.mean(all_mae):.4f} ± {np.std(all_mae):.4f}")
+    print(f"Average MSE: {np.mean(all_mse):.4f} ± {np.std(all_mse):.4f}")
+    print(f"Average R2 : {np.mean(all_r2):.4f} ± {np.std(all_r2):.4f}")
+    print(f"Time: {time.time() - time_star:.2f}s")
+
+
+def EIR_reg():
+    start = time.time()
+    set_seed(42)
+
+    pynvml.nvmlInit()
+    handle = pynvml.nvmlDeviceGetHandleByIndex(0)
+    args = parse_args()
+    all_species =['Cotton/']
+    
+    for i in range(len(all_species)):
+        args.species = all_species[i]
+        device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+        args.device = device  
+        X, Y, nsamples, nsnp, names = load_data(args)
+        for j in range(len(names)):
+            args.phe = names[j]
+            print("starting run " + args.methods + args.species + args.phe)
+            label = Y[:, j]
+            label = np.nan_to_num(label, nan=np.nanmean(label))
+            best_params = Hyperparameter(X, label, nsnp)
+            args.learning_rate = best_params['learning_rate']
+            args.batch_size = best_params['batch_size']
+            args.patience = best_params['patience']
+            start_time = time.time() 
+            torch.cuda.reset_peak_memory_stats()
+            process = psutil.Process(os.getpid())
+
+            run_nested_cv(args, data=X, label=label, nsnp = nsnp, device = args.device)
+
+            elapsed_time = time.time() - start_time
+            print(f"running time: {elapsed_time:.2f} s")
+            print("successfully")
+
+
+if __name__ == '__main__':
+    EIR_reg()

gpbench/method_reg/EIR/EIR_Hyperparameters.py

@@ -0,0 +1,178 @@
+import os
+import time
+import psutil
+import random
+import torch
+import numpy as np
+import optuna
+from sklearn.model_selection import KFold, train_test_split
+from sklearn.preprocessing  import StandardScaler
+from scipy.stats import pearsonr
+from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
+from torch.utils.data  import DataLoader, TensorDataset
+from optuna.exceptions import TrialPruned 
+from .utils.models_locally_connected import LCLModel
+from .utils.common import DataDimensions
+os.environ['CUDA_VISIBLE_DEVICE'] = '0'
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+
+def train_model(model, train_loader, valid_loader, optimizer,criterion, num_epochs, patience, device):
+    model.to(device)
+    best_loss = float('inf')
+    best_state = None
+    trigger_times = 0
+
+    for epoch in range(num_epochs):
+        model.train()
+        train_loss = 0.0
+        for inputs, labels in train_loader:
+            inputs, labels = inputs.to(device), labels.to(device)
+            optimizer.zero_grad()
+            outputs = model(inputs)
+            labels = labels.unsqueeze(1)
+            loss = criterion(outputs, labels)
+            loss.backward()
+            optimizer.step()
+            train_loss += loss.item() * inputs.size(0)
+        # ---------- 验证 ----------
+        model.eval()
+        valid_loss = 0.0
+        with torch.no_grad():
+            for inputs, labels in valid_loader:
+                inputs, labels = inputs.to(device), labels.to(device)
+                outputs = model(inputs)
+                labels = labels.unsqueeze(1)
+                loss = criterion(outputs, labels)
+                valid_loss += loss.item() * inputs.size(0)
+
+        train_loss /= len(train_loader.dataset)
+        valid_loss /= len(valid_loader.dataset)
+
+        # ---------- Early stopping ----------
+        if valid_loss < best_loss:
+            best_loss = valid_loss
+            best_state = model.state_dict()
+            trigger_times = 0
+        else:
+            trigger_times += 1
+            if trigger_times >= patience:
+                print(f"Early stopping at epoch {epoch+1}")
+                break
+
+    # 恢复最佳参数
+    if best_state is not None:
+        model.load_state_dict(best_state)
+    return best_loss
+    
+def predict(model, test_loader, device):
+    model.eval()
+    y_pred = []
+    with torch.no_grad():
+        for inputs, _ in test_loader:
+            outputs = model(inputs)  
+            y_pred.append(outputs.cpu().numpy())
+    y_pred = np.concatenate(y_pred, axis=0)
+    y_pred = np.squeeze(y_pred)
+    return y_pred
+
+def run_nested_cv_with_early_stopping(data, label, nsnp, learning_rate, patience, batch_size,epochs=1000):
+    device = torch.device("cuda:0")
+    print("Starting 10-fold cross-validation...")
+    kf = KFold(n_splits=10, shuffle=True, random_state=42)
+    all_mse, all_mae, all_r2, all_pcc = [], [], [], []
+    
+    for fold, (train_index, test_index) in enumerate(kf.split(data)):
+        print(f"Running fold {fold}...")
+        process = psutil.Process(os.getpid())
+        fold_start_time = time.time()
+
+        X_train, X_test = data[train_index], data[test_index]
+        y_train, y_test = label[train_index], label[test_index]
+
+        X_train_sub, X_valid, y_train_sub, y_valid = train_test_split(X_train, y_train, test_size=0.1, random_state=42)
+        
+        x_train_tensor = torch.from_numpy(X_train_sub).float().to(device)
+        y_train_tensor = torch.from_numpy(y_train_sub).float().to(device)
+        x_valid_tensor = torch.from_numpy(X_valid).float().to(device)
+        y_valid_tensor = torch.from_numpy(y_valid).float().to(device)
+        x_test_tensor = torch.from_numpy(X_test).float().to(device)
+        y_test_tensor = torch.from_numpy(y_test).float().to(device)
+
+        train_data = TensorDataset(x_train_tensor, y_train_tensor)
+        valid_data = TensorDataset(x_valid_tensor, y_valid_tensor)
+        test_data = TensorDataset(x_test_tensor, y_test_tensor)
+
+        train_loader = DataLoader(train_data, batch_size, shuffle=True)
+        valid_loader = DataLoader(valid_data, batch_size, shuffle=False)
+        test_loader = DataLoader(test_data, batch_size, shuffle=False)
+
+        # 初始化模型
+        model = LCLModel(DataDimensions(channels=1, height=nsnp, width=1)).to(device)
+        optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate, weight_decay=1e-4)
+        loss_fn = torch.nn.MSELoss()
+
+        train_model(model, train_loader, valid_loader, optimizer,loss_fn, epochs, patience, device)
+        y_pred = predict(model, test_loader, device)
+
+        # 计算评估指标
+        mse = mean_squared_error(y_test, y_pred)
+        r2 = r2_score(y_test, y_pred)
+        mae = mean_absolute_error(y_test, y_pred)
+        pcc, _ = pearsonr(y_test, y_pred)
+
+        if np.isnan(pcc):
+            print(f"Fold {fold} resulted in NaN PCC, pruning the trial...")
+            raise TrialPruned()
+
+        # 将结果添加到列表中
+        all_mse.append(mse)
+        all_r2.append(r2)
+        all_mae.append(mae)
+        all_pcc.append(pcc)
+
+        # ====== 每折结束时统计 ======
+        fold_time = time.time() - fold_start_time
+        #fold_gpu_mem = #torch.cuda.max_memory_allocated() / 1024**2 if torch.cuda.is_available() else 0
+        fold_cpu_mem = process.memory_info().rss / 1024**2
+        print(f'Fold {fold}:  Corr={pcc:.4f}, MAE={mae:.4f}, MSE={mse:.4f}, R2={r2:.4f}, Time={fold_time:.2f}s, '
+                f'CPU={fold_cpu_mem:.2f}MB')
+        
+    return np.mean(all_pcc) if all_pcc else 0.0
+
+def set_seed(seed=42):
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(seed)
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+
+def Hyperparameter(data, label, nsnp):
+    set_seed(42)
+    
+    def objective(trial):
+        learning_rate = trial.suggest_loguniform("learning_rate", 1e-4,0.1)
+        batch_size = trial.suggest_categorical("batch_size", [32, 64, 128])
+        patience = trial.suggest_int("patience", 10, 100, step=10)
+        try:
+            corr_score = run_nested_cv_with_early_stopping(
+                data=data,
+                label=label,
+                nsnp=nsnp,
+                learning_rate=learning_rate,
+                patience=patience,
+                batch_size=batch_size
+            )
+        
+        except TrialPruned:
+            return float("-inf")
+        return corr_score
+    
+    study = optuna.create_study(direction="maximize")
+    study.optimize(objective, n_trials=20)
+
+    print("最佳参数:", study.best_params)
+    print("successfully")
+    return study.best_params

gpbench/method_reg/EIR/__init__.py

@@ -0,0 +1,5 @@
+from .EIR import EIR_reg
+
+EIR = EIR_reg
+
+__all__ = ["EIR","EIR_reg"]

gpbench/method_reg/EIR/utils/array_output_modules.py

@@ -0,0 +1,97 @@
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Literal, Type
+
+import torch
+from torch import nn
+
+from eir.models.input.array.array_models import al_pre_normalization
+from eir.models.input.array.models_locally_connected import LCLModel, LCLModelConfig
+from eir.models.layers.projection_layers import get_projection_layer
+from eir.models.output.array.output_array_models_cnn import (
+    CNNUpscaleModel,
+    CNNUpscaleModelConfig,
+)
+
+if TYPE_CHECKING:
+    from eir.setup.input_setup_modules.common import DataDimensions
+
+al_array_model_types = Literal["lcl", "cnn"]
+al_output_array_model_classes = Type[LCLModel] | Type[CNNUpscaleModel]
+al_output_array_models = LCLModel | CNNUpscaleModel
+al_output_array_model_config_classes = (
+    Type["LCLOutputModelConfig"] | Type[CNNUpscaleModelConfig]
+)
+
+
+@dataclass
+class LCLOutputModelConfig(LCLModelConfig):
+    cutoff: int | Literal["auto"] = "auto"
+
+
+@dataclass
+class ArrayOutputModuleConfig:
+
+    """
+    :param model_type:
+         Which type of image model to use.
+
+    :param model_init_config:
+          Configuration used to initialise model.
+    """
+
+    model_type: al_array_model_types
+    model_init_config: LCLOutputModelConfig
+    pre_normalization: al_pre_normalization = None
+
+
+class ArrayOutputWrapperModule(nn.Module):
+    def __init__(
+        self,
+        feature_extractor: al_output_array_models,
+        output_name: str,
+        target_data_dimensions: "DataDimensions",
+    ):
+        super().__init__()
+        self.feature_extractor = feature_extractor
+        self.output_name = output_name
+        self.data_dimensions = target_data_dimensions
+
+        self.target_width = self.data_dimensions.num_elements()
+        self.target_shape = self.data_dimensions.full_shape()
+
+        diff_tolerance = get_diff_tolerance(num_target_elements=self.target_width)
+
+        self.projection_head = get_projection_layer(
+            input_dimension=self.feature_extractor.num_out_features,
+            target_dimension=self.target_width,
+            projection_layer_type="lcl_residual",
+            lcl_diff_tolerance=diff_tolerance,
+        )
+
+    def forward(self, x: torch.Tensor) -> dict[str, torch.Tensor]:
+        out = self.feature_extractor(x)
+
+        out = out.reshape(out.shape[0], -1)
+        out = self.projection_head(out)
+
+        out = out[:, : self.target_width]
+
+        out = out.reshape(-1, *self.target_shape)
+
+        return {self.output_name: out}
+
+
+def get_diff_tolerance(num_target_elements: int) -> int:
+    return int(0.001 * num_target_elements)
+
+
+def get_array_output_module(
+    feature_extractor: al_output_array_models,
+    output_name: str,
+    target_data_dimensions: "DataDimensions",
+) -> ArrayOutputWrapperModule:
+    return ArrayOutputWrapperModule(
+        feature_extractor=feature_extractor,
+        output_name=output_name,
+        target_data_dimensions=target_data_dimensions,
+    )

gpbench/method_reg/EIR/utils/common.py

@@ -0,0 +1,65 @@
+from dataclasses import dataclass
+from pathlib import Path
+from typing import List, Optional, Tuple
+
+import numpy as np
+
+# from eir.data_load.data_source_modules.deeplake_ops import (
+#     get_deeplake_input_source_iterable,
+#     is_deeplake_dataset,
+#     load_deeplake_dataset,
+# )
+# from eir.data_load.label_setup import get_file_path_iterator
+
+
+@dataclass
+class DataDimensions:
+    channels: int
+    height: int
+    width: int
+    extra_dims: tuple[int, ...] = tuple()
+
+    def num_elements(self) -> int:
+        base = self.channels * self.height * self.width
+        return int(base * np.prod(self.extra_dims))
+
+    def full_shape(self) -> Tuple[int, ...]:
+        return (self.channels, self.height, self.width) + self.extra_dims
+
+
+# def get_data_dimension_from_data_source(
+#     data_source: Path,
+#     deeplake_inner_key: Optional[str] = None,
+# ) -> DataDimensions:
+#     """
+#     TODO: Make more dynamic / robust. Also weird to say "width" for a 1D vector.
+#     """
+#
+#     if is_deeplake_dataset(data_source=str(data_source)):
+#         assert deeplake_inner_key is not None, data_source
+#         deeplake_ds = load_deeplake_dataset(data_source=str(data_source))
+#         deeplake_iter = get_deeplake_input_source_iterable(
+#             deeplake_dataset=deeplake_ds, inner_key=deeplake_inner_key
+#         )
+#         shape = next(deeplake_iter).shape
+#     else:
+#         iterator = get_file_path_iterator(data_source=data_source)
+#         path = next(iterator)
+#         shape = np.load(file=path).shape
+#
+#     extra_dims: tuple[int, ...] = tuple()
+#     if len(shape) == 1:
+#         channels, height, width = 1, 1, shape[0]
+#     elif len(shape) == 2:
+#         channels, height, width = 1, shape[0], shape[1]
+#     elif len(shape) == 3:
+#         channels, height, width = shape
+#     else:
+#         channels, height, width = shape[0], shape[1], shape[2]
+#         extra_dims = shape[3:]
+#
+#     return DataDimensions(
+#         channels=channels, height=height, width=width, extra_dims=extra_dims
+#     )
+
+