gpbench 1.0.0__py3-none-any.whl

gpbench/method_reg/CropARNet/base_CropARNet.py

@@ -0,0 +1,137 @@
+import torch
+import torch.nn as nn
+import numpy as np
+
+config = {
+    "batch_size": 64,
+    "weights_units": [64, 32],
+    "regressor_units": [64, 32],
+    "dropout": 0.3,
+}
+
+
+class SimpleSNPModel(nn.Module):
+    def __init__(self, num_snps):
+        super().__init__()
+        try:
+            self.config = config
+            if not isinstance(num_snps, int) or num_snps <= 0:
+                raise ValueError(f"num_snps must be positive integer, got {num_snps}")
+                
+            self.attention = self._build_attention_module(num_snps)
+            self.regressor = self._build_regressor_module(num_snps)
+        except Exception as e:
+            raise ValueError(f"Model initialization failed: {str(e)}")
+
+    def _build_attention_module(self, num_snps):
+        """Build attention module with error checking"""
+        try:
+            layers = []
+            prev_size = num_snps
+            for i, h_size in enumerate(self.config['weights_units']):
+                if not isinstance(h_size, int) or h_size <= 0:
+                    raise ValueError(f"Invalid hidden size {h_size} in attention layer {i}")
+                layers.append(nn.Linear(prev_size, h_size))
+                if i < len(self.config['weights_units']) - 1:
+                    layers.append(nn.GELU())
+                prev_size = h_size
+            layers.append(nn.Linear(prev_size, num_snps))
+            layers.append(nn.Sigmoid())
+            return nn.Sequential(*layers)
+        except Exception as e:
+            raise ValueError(f"Attention module construction failed: {str(e)}")
+
+    def _build_regressor_module(self, num_snps):
+        """Build regressor module with error checking"""
+        try:
+            layers = []
+            prev_size = num_snps
+            for i, h_size in enumerate(self.config['regressor_units']):
+                if not isinstance(h_size, int) or h_size <= 0:
+                    raise ValueError(f"Invalid hidden size {h_size} in regressor layer {i}")
+                layers.append(nn.Linear(prev_size, h_size))
+                if i < len(self.config['regressor_units']) - 1:
+                    layers.append(nn.LayerNorm(h_size))
+                    layers.append(nn.GELU())
+                    layers.append(nn.Dropout(self.config['dropout']))
+                prev_size = h_size
+            layers.append(nn.Linear(prev_size, 1))
+            return nn.Sequential(*layers)
+        except Exception as e:
+            raise ValueError(f"Regressor module construction failed: {str(e)}")
+
+    def forward(self, x):
+        """Forward pass with dimension checking"""
+        try:
+            if x.dim() != 2:
+                raise ValueError(f"Input must be 2D tensor, got {x.dim()}D")
+                
+            pre_sigmoid_weights = self.attention[:-1](x) 
+            att_weights = self.attention(x) 
+            weighted = x * att_weights + x  # Residual connection
+            return self.regressor(weighted).squeeze(), pre_sigmoid_weights
+        except Exception as e:
+            raise RuntimeError(f"Forward pass failed: {str(e)}")
+        
+
+    def train_model(self, train_loader, valid_loader, num_epochs, learning_rate, weight_decay, patience, device):
+        optimizer = torch.optim.AdamW(self.parameters(), lr=learning_rate, weight_decay=weight_decay)
+        criterion = nn.MSELoss()
+        self.to(device)
+
+        best_loss = float('inf')
+        best_state = None
+        trigger_times = 0
+
+        for epoch in range(num_epochs):
+            self.train()
+            train_loss = 0.0
+            for inputs, labels in train_loader:
+                inputs, labels = inputs.to(device), labels.to(device)
+                optimizer.zero_grad()
+                outputs, _ = self(inputs)
+                labels = labels
+                loss = criterion(outputs, labels)
+                loss.backward()
+                optimizer.step()
+                train_loss += loss.item() * inputs.size(0)
+
+            self.eval()
+            valid_loss = 0.0
+            with torch.no_grad():
+                for inputs, labels in valid_loader:
+                    inputs, labels = inputs.to(device), labels.to(device)
+                    outputs,_ = self(inputs)
+                    labels = labels
+                    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 = self.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:
+            self.load_state_dict(best_state)
+        return best_loss
+        
+    def predict(self, test_loader):
+        self.eval()
+        y_pred = []
+        with torch.no_grad():
+            for inputs, _ in test_loader:
+                outputs,_ = self(inputs)  
+                y_pred.append(outputs.cpu().numpy())
+        y_pred = np.concatenate(y_pred, axis=0)
+        y_pred = np.squeeze(y_pred)
+        return y_pred
+

gpbench/method_reg/Cropformer/Cropformer.py

@@ -0,0 +1,313 @@
+import argparse
+import random
+import torch
+import numpy as np
+import pandas as pd
+import torch.nn as nn
+import time, psutil, os 
+import torch.optim as optim 
+from torch.optim import Adam
+from torch.nn import MSELoss
+from lightning.pytorch import LightningModule
+from sklearn.model_selection import KFold
+from torch.utils.data  import DataLoader, TensorDataset 
+from sklearn.preprocessing  import StandardScaler
+from sklearn.metrics  import mean_absolute_error, mean_squared_error, r2_score
+from . import Cropformer_Hyperparameters
+import pynvml
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Argument parser")
+    parser.add_argument('--methods', type=str, default='Cropformer/', help='Random seed')
+    parser.add_argument('--species', type=str, default='Chickpea/GSTP012/', help='Dataset name')
+    parser.add_argument('--phe', type=str, default='', help='Dataset name')
+    parser.add_argument('--data_dir', type=str, default='../../data/', help='Path to data directory')
+    parser.add_argument('--result_dir', type=str, default='result/', help='Path to result directory')
+
+    parser.add_argument('--lr', type=float, default=0.01,help='Learning rate')
+    parser.add_argument('--num_head', type=int, default=1, help='Number of attention heads')
+    parser.add_argument('--dropout', type=float, default=0.5, help='Dropout probability')
+    parser.add_argument('--batch_size', type=int, default=32, help='Batch size')
+    parser.add_argument('--hidden_dim', type=int, default=64, help='Hidden dimension')
+    parser.add_argument('--kernel_size', type=int, default=3, help='Kernel size')
+    parser.add_argument('--patience', type=int, default=5, 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
+
+class LayerNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-12):
+        super(LayerNorm, self).__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.bias = nn.Parameter(torch.zeros(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, x):
+        u = x.mean(-1, keepdim=True)
+        s = (x - u).pow(2).mean(-1, keepdim=True)
+        x = (x - u) / torch.sqrt(s + self.variance_epsilon)
+        return self.weight * x + self.bias
+        
+class SelfAttention(LightningModule):
+    def __init__(self, num_attention_heads, input_size, hidden_size, output_dim=1, kernel_size=3,
+                 hidden_dropout_prob=0.5, attention_probs_dropout_prob=0.5, learning_rate=0.001):
+        super(SelfAttention, self).__init__()
+        self.num_attention_heads = num_attention_heads
+        self.attention_head_size = int(hidden_size / num_attention_heads)
+        self.all_head_size = hidden_size
+
+        self.query = torch.nn.Linear(input_size, self.all_head_size)
+        self.key = torch.nn.Linear(input_size, self.all_head_size)
+        self.value = torch.nn.Linear(input_size, self.all_head_size)
+
+        self.attn_dropout = torch.nn.Dropout(attention_probs_dropout_prob)
+        self.out_dropout = torch.nn.Dropout(hidden_dropout_prob)
+        self.dense = torch.nn.Linear(hidden_size, input_size)
+        self.LayerNorm = torch.nn.LayerNorm(input_size, eps=1e-12)
+        self.relu = torch.nn.ReLU()
+        self.out = torch.nn.Linear(input_size, output_dim)
+        self.cnn = torch.nn.Conv1d(1, 1, kernel_size, stride=1, padding=1)
+
+        self.learning_rate = learning_rate
+        self.loss_fn = MSELoss()
+
+    def forward(self, input_tensor):
+        input_tensor = input_tensor.to(self.device)
+        self.cnn = self.cnn.to(self.device)
+
+        cnn_hidden = self.cnn(input_tensor.view(input_tensor.size(0), 1, -1))
+        input_tensor = cnn_hidden
+        mixed_query_layer = self.query(input_tensor)
+        mixed_key_layer = self.key(input_tensor)
+        mixed_value_layer = self.value(input_tensor)
+
+        query_layer = mixed_query_layer
+        key_layer = mixed_key_layer
+        value_layer = mixed_value_layer
+
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+        attention_scores = attention_scores / np.sqrt(self.attention_head_size)
+        attention_probs = torch.nn.Softmax(dim=-1)(attention_scores)
+        attention_probs = self.attn_dropout(attention_probs)
+
+        context_layer = torch.matmul(attention_probs, value_layer)
+        hidden_states = self.dense(context_layer)
+        hidden_states = self.out_dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        output = self.out(self.relu(hidden_states.view(hidden_states.size(0), -1)))
+        return output
+
+    def training_step(self, batch, batch_idx):
+        x, y = batch
+        y_pred = self(x)
+        loss = self.loss_fn(y_pred, y)
+        return loss
+
+    def validation_step(self, batch, batch_idx):
+        x, y = batch
+        y_pred = self(x)
+        val_loss = self.loss_fn(y_pred, y)
+        return val_loss
+
+    def configure_optimizers(self):
+        return Adam(self.parameters(), lr=self.learning_rate)
+
+class EarlyStopping:
+    def __init__(self, patience=10, delta=0):
+        self.patience = patience
+        self.delta = delta
+        self.best_score = None
+        self.counter = 0
+        self.early_stop = False
+
+    def __call__(self, score):
+        if self.best_score is None:
+            self.best_score = score
+        elif score < self.best_score + self.delta:
+            self.counter += 1
+            if self.counter >= self.patience:
+                self.early_stop = True
+        else:
+            self.best_score = score
+            self.counter = 0
+
+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 run_nested_cv_with_early_stopping(args, data, label, outer_cv, learning_rate, batch_size, hidden_dim,
+                                      output_dim, kernel_size, patience, DEVICE):
+    result_dir = os.path.join(args.result_dir, args.methods + args.species + args.phe)
+    os.makedirs(result_dir, exist_ok=True)
+    best_corr_coefs = []
+    best_maes = []
+    best_r2s = []
+    best_mses = []
+
+    time_star = time.time()
+    process = psutil.Process(os.getpid())
+    for fold, (train_idx, test_idx) in enumerate(outer_cv.split(data)):
+        fold_start_time = time.time()
+        if torch.cuda.is_available():
+            torch.cuda.reset_peak_memory_stats()
+        process = psutil.Process(os.getpid())
+
+        x_train, x_test = data[train_idx], data[test_idx]
+        y_train, y_test = label[train_idx], label[test_idx]
+        
+        num_attention_heads = args.num_head
+        attention_probs_dropout_prob = args.dropout
+        hidden_dropout_prob = 0.5
+
+        model = SelfAttention(num_attention_heads, x_train.shape[1], hidden_dim, output_dim,
+                                hidden_dropout_prob=hidden_dropout_prob, kernel_size=kernel_size,
+                                attention_probs_dropout_prob=attention_probs_dropout_prob).to(DEVICE)
+        optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
+        loss_function = torch.nn.MSELoss()
+        scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='max', factor=0.1, patience=10)
+
+        scaler = StandardScaler()
+        x_train = scaler.fit_transform(x_train)
+        x_test = scaler.transform(x_test)
+
+        x_train_tensor = torch.from_numpy(x_train).float().to(DEVICE)
+        y_train_tensor = torch.from_numpy(y_train).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)
+        test_data = TensorDataset(x_test_tensor, y_test_tensor)
+
+        train_loader = DataLoader(train_data, batch_size, shuffle=True)
+        test_loader = DataLoader(test_data, batch_size, shuffle=False)
+
+        early_stopping = EarlyStopping(patience=patience)
+        best_corr_coef = -float('inf')
+        best_mae = float('inf')
+        best_mse = float('inf')
+        best_r2 = -float('inf')
+        for epoch in range(100):
+            model.train()
+            for x_batch, y_batch in train_loader:
+                optimizer.zero_grad()
+                y_pred = model(x_batch)
+                loss = loss_function(y_pred, y_batch.reshape(-1, 1))
+                loss.backward()
+                optimizer.step()
+
+            model.eval()
+            y_test_preds, y_test_trues = [], []
+            
+            with torch.no_grad():
+                for x_batch, y_batch in test_loader:
+                    y_test_pred = model(x_batch)
+                    y_test_preds.extend(y_test_pred.cpu().numpy().reshape(-1).tolist())
+                    y_test_trues.extend(y_batch.cpu().numpy().reshape(-1).tolist())
+
+            corr_coef = np.corrcoef(y_test_preds, y_test_trues)[0, 1]
+            mae = mean_absolute_error(np.array(y_test_trues), np.array(y_test_preds))
+            mse = mean_squared_error(np.array(y_test_trues), np.array(y_test_preds))
+            r2 = r2_score(np.array(y_test_trues), np.array(y_test_preds))
+            scheduler.step(corr_coef)
+
+            
+            if corr_coef > best_corr_coef:
+                best_mae = mae
+                best_corr_coef = corr_coef
+                best_mse = mse
+                best_r2 = r2
+
+            early_stopping(corr_coef)
+            if early_stopping.early_stop:
+                print(f"Early stopping at epoch {epoch + 1}")
+                break
+        
+        best_corr_coefs.append(best_corr_coef)
+        best_maes.append(best_mae)
+        best_mses.append(best_mse)
+        best_r2s.append(best_r2)
+
+        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 + 1}:  Corr={best_corr_coef:.4f}, MAE={best_mae:.4f}, MSE={best_mse:.4f}, R2={best_r2:.4f}, Time={fold_time:.2f}s, '
+              f'GPU={fold_gpu_mem:.2f}MB, CPU={fold_cpu_mem:.2f}MB')
+        
+        results_df = pd.DataFrame({'Y_test': y_test_trues, 'Y_pred': y_test_preds})
+        results_df.to_csv(os.path.join(result_dir, f"fold{fold}.csv"), index=False)
+
+    print("==== Final Results ====")
+    print(f"Corr: {np.mean(best_corr_coefs):.4f} ± {np.std(best_corr_coefs):.4f}")
+    print(f"MAE: {np.mean(best_maes):.4f} ± {np.std(best_maes):.4f}")
+    print(f"MSE: {np.mean(best_mses):.4f} ± {np.std(best_mses):.4f}")
+    print(f"R2 : {np.mean(best_r2s):.4f} ± {np.std(best_r2s):.4f}")
+    print(f"Time: {time.time() - time_star:.2f}s")
+
+
+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 Cropformer_reg():
+    set_seed(42)
+    pynvml.nvmlInit()
+    handle = pynvml.nvmlDeviceGetHandleByIndex(0)
+    device = torch.device("cuda:0")
+    args = parse_args()
+    all_species =['Cotton/']
+    for i in range(len(all_species)):
+        args.species = all_species[i]
+        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 = Cropformer_Hyperparameters.Hyperparameter(X, label)
+            args.lr = best_params['learning_rate']
+            args.num_head = best_params['heads']
+            args.dropout = best_params['dropout']
+            args.batch_size = best_params['batch_size']
+
+            outer_cv = KFold(n_splits=10, shuffle=True, random_state=42)
+            
+            start_time = time.time()
+            torch.cuda.reset_peak_memory_stats()
+            process = psutil.Process(os.getpid())
+
+            run_nested_cv_with_early_stopping(args, 
+                                            data=X,
+                                            label=label,
+                                            outer_cv=outer_cv,
+                                            learning_rate= args.lr,
+                                            batch_size=args.batch_size,
+                                            hidden_dim=args.hidden_dim,
+                                            output_dim=1,
+                                            kernel_size=3,
+                                            patience=args.patience,
+                                            DEVICE='cuda:0' if torch.cuda.is_available() else 'cpu')
+            
+            elapsed_time = time.time() - start_time
+            print(f"running time: {elapsed_time:.2f} s")
+            print("successfully")
+
+
+if __name__ == '__main__':
+    Cropformer_reg()