junshan-kit 2.5.1__py2.py3-none-any.whl → 2.8.5__py2.py3-none-any.whl

junshan_kit/BenchmarkFunctions.py

@@ -0,0 +1,7 @@
+
+
+def rosenbrock(x, a=1.0, b=100.0):
+    # Optimal value: (a, a^2)
+    return (a - x[0])**2 + b * (x[1] - x[0]**2)**2
+
+

junshan_kit/Check_Info.py

@@ -0,0 +1,44 @@
+"""
+----------------------------------------------------------------------
+>>> Author       : Junshan Yin  
+>>> Last Updated : 2025-11-22
+----------------------------------------------------------------------
+"""
+
+from junshan_kit import ModelsHub
+
+def check_args(self, args, parser, allowed_models, allowed_optimizers, allowed_datasets):
+    # Parse and validate each train_group
+    for cfg in args.train:
+        try:
+            model, dataset, optimizer = cfg.split("-")
+
+            if model not in allowed_models:
+                parser.error(f"Invalid model '{model}'. Choose from {allowed_models}")
+            if optimizer not in allowed_optimizers:
+                parser.error(f"Invalid optimizer '{optimizer}'. Choose from {allowed_optimizers}")
+            if dataset not in allowed_datasets:
+                parser.error(f"Invalid dataset '{dataset}'. Choose from {allowed_datasets}")
+
+        except ValueError:
+            parser.error(f"Invalid format '{cfg}'. Use model-dataset-optimizer")
+
+    for cfg in args.train:
+        model_name, dataset_name, optimizer_name = cfg.split("-")
+        try:
+            f = getattr(ModelsHub, f"Build_{args.model_name_mapping[model_name]}_{args.data_name_mapping[dataset_name]}")
+
+        except:
+            print(getattr(ModelsHub, f"Build_{args.model_name_mapping[model_name]}_{args.data_name_mapping[dataset_name]}"))
+            assert False
+
+def check_subset_info(self, args, parser):
+        total = sum(args.subset)
+        if args.subset[0]>1:
+            # CHECK
+            for i in args.subset:
+                if i < 1:
+                    parser.error(f"Invalid --subset {args.subset}: The number of subdata must > 1")    
+        else:
+            if abs(total - 1.0) != 0.0:  
+                parser.error(f"Invalid --subset {args.subset}: the values must sum to 1.0 (current sum = {total:.6f}))")

junshan_kit/DataHub.py

@@ -1,41 +1,106 @@
+"""
+----------------------------------------------------------------------
+>>> Author       : Junshan Yin  
+>>> Last Updated : 2025-10-28
+----------------------------------------------------------------------
+"""
+
 import torchvision, torch
 import torchvision.transforms as transforms
 import pandas as pd
+from torch.utils.data import random_split, Subset
 
-from junshan_kit import DataSets, DataProcessor
+from junshan_kit import DataSets, DataProcessor, ParametersHub
 
+def Adult_Income_Prediction(Paras):
 
+    df = DataSets.adult_income_prediction() 
+    transform = {
+        "train_size": 0.7,
+        "normalization": True
+    }
+    label_col='income'
 
+    train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
 
+    return train_dataset, test_dataset, transform
 
 
+def Credit_Card_Fraud_Detection(Paras):
+    df = DataSets.credit_card_fraud_detection()
+    transform = {
+        "train_size": 0.7,
+        "normalization": True
+    }
+    label_col='Class'
 
+    train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
 
+    return train_dataset, test_dataset, transform
 
+def Diabetes_Health_Indicators(Paras):
+    df = DataSets.diabetes_health_indicators()
+    transform = {
+        "train_size": 0.7,
+        "normalization": True
+    }
+    label_col='diagnosed_diabetes'
 
+    train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
 
-def Adult_Income_Prediction(Paras):
+    return train_dataset, test_dataset, transform
 
-    df = DataSets.adult_income_prediction() 
+def Electric_Vehicle_Population(Paras): 
+    df = DataSets.electric_vehicle_population()
     transform = {
         "train_size": 0.7,
         "normalization": True
     }
-    label_col='income'
+    label_col='Electric Vehicle Type'
+    train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
+
+    return train_dataset, test_dataset, transform
 
+def Global_House_Purchase(Paras): 
+    df = DataSets.global_house_purchase()
+    transform = {
+        "train_size": 0.7,
+        "normalization": True
+    }
+    label_col='decision'
     train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
 
     return train_dataset, test_dataset, transform
 
+def Health_Lifestyle(Paras): 
+    df = DataSets.health_lifestyle()
+    transform = {
+        "train_size": 0.7,
+        "normalization": True
+    }
+    label_col='disease_risk'
+    train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
 
-def Credit_Card_Fraud_Detection(Paras):
-    df = DataSets.credit_card_fraud_detection()
+    return train_dataset, test_dataset, transform
+
+def Homesite_Quote_Conversion(Paras): 
+    df = DataSets.Homesite_Quote_Conversion()
     transform = {
         "train_size": 0.7,
         "normalization": True
     }
-    label_col='Class'
+    label_col='QuoteConversion_Flag'
+    train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
+
+    return train_dataset, test_dataset, transform
 
+def TN_Weather_2020_2025(Paras): 
+    df = DataSets.TamilNadu_weather_2020_2025()
+    transform = {
+        "train_size": 0.7,
+        "normalization": True
+    }
+    label_col='rain_tomorrow'
     train_dataset, test_dataset, transform = DataProcessor.Pandas_TO_Torch(df, label_col).to_torch(transform, Paras)
 
     return train_dataset, test_dataset, transform
@@ -101,6 +166,7 @@ def CIFAR100(Paras, model_name):
         download=True,
         transform=transform
     )
+
     if Paras["model_type"][model_name] == "binary":
         train_mask = (torch.tensor(train_dataset.targets) == 0) | (torch.tensor(train_dataset.targets) == 1)
         test_mask = (torch.tensor(test_dataset.targets) == 0) | (torch.tensor(test_dataset.targets) == 1)
@@ -111,4 +177,38 @@ def CIFAR100(Paras, model_name):
         train_dataset = torch.utils.data.Subset(train_dataset, train_indices.tolist())
         test_dataset = torch.utils.data.Subset(test_dataset, test_indices.tolist())
 
-    return train_dataset, test_dataset, transform
+    return train_dataset, test_dataset, transform
+
+
+def Caltech101_Resize_32(Paras, train_ratio=0.7, split=True):
+
+    transform = transforms.Compose([
+        # transforms.Lambda(convert_to_rgb),  
+        transforms.Grayscale(num_output_channels=3),
+        transforms.Resize((32, 32)),
+        transforms.ToTensor(),
+        transforms.Normalize(mean=[0.485, 0.456, 0.406],  
+                            std=[0.229, 0.224, 0.225])
+    ])
+    
+    full_dataset = torchvision.datasets.Caltech101(
+        root='./exp_data/Caltech101',
+        download=True,
+        transform=transform
+    )
+
+    if split:
+        total_size = len(full_dataset)
+        train_size = int(train_ratio * total_size)
+        test_size = total_size - train_size
+
+        train_dataset, test_dataset = random_split(full_dataset, [train_size, test_size])
+
+    else:
+        train_dataset = full_dataset
+        # Empty test dataset, keep the structure consistent
+        test_dataset = Subset(full_dataset, [])  
+
+    return train_dataset, test_dataset, transform
+
+# <caltech101_Resize_32>

junshan_kit/DataProcessor.py

@@ -6,8 +6,13 @@
 """
 
 import pandas as pd
-import torch
+import numpy as np
+import torch, bz2
+from typing import Optional
+from torch.utils.data import random_split, Subset
+from sklearn.datasets import load_svmlight_file
 from sklearn.preprocessing import StandardScaler
+from junshan_kit import ParametersHub
 
 
 class CSV_TO_Pandas:
@@ -181,7 +186,7 @@ class CSV_TO_Pandas:
         if time_info is not None:
             df = self._trans_time_fea(df, time_info)
 
-        # Step 3: Map target label (to -1 and +1)
+        # Step 3: Map target label (to 0 and +1)
         df[label_col] = df[label_col].map(label_map)
 
         # Step 4: Encode categorical features (exclude label column)
@@ -200,7 +205,7 @@ class CSV_TO_Pandas:
         # print info
         if print_info:
             pos_count = (df[label_col] == 1).sum()
-            neg_count = (df[label_col] == -1).sum()
+            neg_count = (df[label_col] == 0).sum()
 
             # Step 6: Print dataset information
             print("\n" + "=" * 80)
@@ -214,8 +219,8 @@ class CSV_TO_Pandas:
             print(
                 f"{'Dropping NaN & non-feature cols:':<40} {m_encoded} rows x {n_encoded} cols"
             )
-            print(f"{'Positive samples (+1):':<40} {pos_count}")
-            print(f"{'Negative samples (-1):':<40} {neg_count}")
+            print(f"{'Positive samples (1):':<40} {pos_count}")
+            print(f"{'Negative samples (0):':<40} {neg_count}")
             print(
                 f"{'Size after one-hot encoding:':<40} {m_cleaned} rows x {n_cleaned} cols"
             )
@@ -252,7 +257,7 @@ from torch.utils.data import Dataset
 class Pandas_TO_Torch(Dataset):
 
     def __init__(self, df: pd.DataFrame, 
-                 label_col: str, 
+                label_col: str, 
                 ):
         self.df = df
         self.label_col = label_col
@@ -316,8 +321,6 @@ class bz2_To_Numpy:
 
 
 
-    
-
 class StepByStep:
     def __init__(self):
         pass
@@ -332,3 +335,125 @@ class StepByStep:
             )
 
 
+class LibSVMDataset_bz2(Dataset):
+    def __init__(self, path, data_name = None, Paras = None):
+        with bz2.open(path, 'rb') as f:
+            X, y = load_svmlight_file(f) # type: ignore
+
+        self.X, self.path = X, path
+        
+        y = np.asanyarray(y)
+
+        if data_name is not None:
+            data_name = data_name.lower()
+
+            # Binary classification, with the label -1/1
+            if data_name in ["rcv1"]:
+                y = (y > 0).astype(int)  # Convert to 0/1
+            
+            # Multi-category, labels usually start with 1
+            elif data_name in [""]:  
+                y = y - 1  # Start with 0
+
+        else:
+            # Default policy: Try to avoid CrossEntropyLoss errors
+            if np.min(y) < 0:  # e.g. [-1, 1]
+                y = (y > 0).astype(int)
+            elif np.min(y) >= 1:
+                y = y - 1
+
+        self.y = y
+
+    def __len__(self):
+        return self.X.shape[0]
+
+    def __getitem__(self, idx):
+        xi = torch.tensor(self.X.getrow(idx).toarray(), dtype=torch.float32).squeeze(0)
+        yi = torch.tensor(self.y[idx], dtype=torch.float32)
+        return xi, yi
+    
+    def __repr__(self):
+        num_samples = len(self.y)
+        num_features = self.X.shape[1]
+        num_classes = len(np.unique(self.y))
+        return (f"LibSVMDataset_bz2(\n"
+                f"  num_samples = {num_samples},\n"
+                f"  num_features = {num_features},\n"
+                f"  num_classes = {num_classes}\n"
+                f"  path = {self.path}\n"
+                f")")
+    
+def get_libsvm_bz2_data(train_path, test_path, data_name, Paras, split = True):
+    
+    transform = "-1 → 0 for binary, y-1 for multi-class"
+    train_data = LibSVMDataset_bz2(train_path)
+
+    if data_name in ["Duke", "Ijcnn", "RCV1"]:
+        test_data = LibSVMDataset_bz2(test_path)
+        split = False
+    else:
+        test_data = Subset(train_data, [])
+        
+    
+    if split:
+        total_size = len(train_data)
+        train_size = int(Paras["train_ratio"] * total_size)
+        test_size = total_size - train_size
+
+        train_dataset, test_dataset = random_split(train_data, [train_size, test_size])
+
+    else:
+        train_dataset = train_data
+        # # Empty test dataset, keep the structure consistent
+        # test_dataset = Subset(train_data, []) 
+        test_dataset = test_data
+
+    # print(test_dataset) 
+    # assert False
+
+    return train_dataset, test_dataset, transform
+
+
+def subset(dataset, ratio_or_num, seed=None) -> Subset:
+    """
+    Randomly sample a subset from a dataset.
+
+    Parameters
+    ----------
+    dataset : torch.utils.data.Dataset
+        The dataset to sample from.
+    ratio_or_num : float or int
+        If float in (0, 1], treated as sampling ratio.
+        Otherwise, treated as absolute number of samples.
+    seed : int, optional
+        Random seed for reproducibility.
+
+    Returns
+    -------
+    torch.utils.data.Subset
+        A randomly sampled subset of the dataset.
+    """
+
+    if ratio_or_num < 0:
+        raise ValueError(f"ratio_or_num must be non-negative, got {ratio_or_num}")
+
+    dataset_len = len(dataset)
+
+    # Determine number of samples
+    if isinstance(ratio_or_num, float) and 0 < ratio_or_num <= 1:
+        num = max(1, int(round(dataset_len * ratio_or_num)))
+    else:
+        num = int(ratio_or_num)
+
+    # Clamp to valid range
+    num = min(max(num, 1), dataset_len)
+
+    # Create and seed generator
+    generator = torch.Generator()
+    if seed is not None:
+        generator.manual_seed(seed)
+
+    # Random sampling
+    indices = torch.randperm(dataset_len, generator=generator)[:num].tolist()
+
+    return Subset(dataset, indices)

junshan_kit/DataSets.py

@@ -147,12 +147,12 @@ def _run(csv_path, data_name, data_type, drop_cols, label_col, label_map, print_
 ----------------------------------------------------------------------
 """
 
-def credit_card_fraud_detection(data_name = "Credit Card Fraud Detection", print_info = False, export_csv=False, drop_cols = []):
+def credit_card_fraud_detection(data_name = "Credit_Card_Fraud_Detection", print_info = False, export_csv=False, drop_cols = []):
 
     data_type = "binary"
-    csv_path = f'./exp_data/{data_name}/creditcard.csv'
+    csv_path = f'exp_data/{data_name}/creditcard.csv'
     label_col = 'Class' 
-    label_map = {0: -1, 1: 1}
+    label_map = {0: 0, 1: 1}
     
 
     df = _run(csv_path, data_name, data_type, drop_cols, label_col, label_map, print_info, export_csv=export_csv)
@@ -161,26 +161,26 @@ def credit_card_fraud_detection(data_name = "Credit Card Fraud Detection", print
     return df
 
 
-def diabetes_health_indicators(data_name = "Diabetes Health Indicators", print_info = False, export_csv = False, drop_cols = [], Standard = False):
+def diabetes_health_indicators(data_name = "Diabetes_Health_Indicators", print_info = False, export_csv = False, drop_cols = [], Standard = False):
     data_type = "binary"
-    csv_path = f'./exp_data/{data_name}/diabetes_dataset.csv'
+    csv_path = f'exp_data/{data_name}/diabetes_dataset.csv'
     label_col = 'diagnosed_diabetes'
-    label_map = {0: -1, 1: 1}
+    label_map = {0: 0, 1: 1}
     
     df = _run(csv_path, data_name, data_type, drop_cols, label_col, label_map, print_info, export_csv=export_csv)
 
     return df
 
 
-def electric_vehicle_population(data_name = "Electric Vehicle Population", print_info = False, export_csv = False, drop_cols = ['VIN (1-10)', 'DOL Vehicle ID', 'Vehicle Location'], Standard = False):
+def electric_vehicle_population(data_name = "Electric_Vehicle_Population", print_info = False, export_csv = False, drop_cols = ['VIN (1-10)', 'DOL Vehicle ID', 'Vehicle Location'], Standard = False):
 
     data_type = "binary"
-    csv_path = f'./exp_data/{data_name}/Electric_Vehicle_Population_Data.csv'
+    csv_path = f'exp_data/{data_name}/Electric_Vehicle_Population_Data.csv'
     # drop_cols = ['VIN (1-10)', 'DOL Vehicle ID', 'Vehicle Location']
     label_col = 'Electric Vehicle Type'
     label_map = {
     'Battery Electric Vehicle (BEV)': 1,
-    'Plug-in Hybrid Electric Vehicle (PHEV)': -1
+    'Plug-in Hybrid Electric Vehicle (PHEV)': 0
     }
     
 
@@ -188,12 +188,12 @@ def electric_vehicle_population(data_name = "Electric Vehicle Population", print
 
     return df
 
-def global_house_purchase(data_name = "Global House Purchase", print_info = False, export_csv = False, drop_cols = ['property_id'], Standard =False):
+def global_house_purchase(data_name = "Global_House_Purchase", print_info = False, export_csv = False, drop_cols = ['property_id'], Standard =False):
 
     data_type = "binary"
-    csv_path = f'./exp_data/{data_name}/global_house_purchase_dataset.csv'
+    csv_path = f'exp_data/{data_name}/global_house_purchase_dataset.csv'
     label_col = 'decision'
-    label_map = {0: -1, 1: 1}
+    label_map = {0: 0, 1: 1}
     
 
     df = _run(csv_path, data_name, data_type, drop_cols, label_col, label_map, print_info, export_csv=export_csv)
@@ -201,13 +201,13 @@ def global_house_purchase(data_name = "Global House Purchase", print_info = Fals
     return df
 
 
-def health_lifestyle(data_name = "Health Lifestyle", print_info = False, export_csv = False, drop_cols = ['id'], Standard =False):
+def health_lifestyle(data_name = "Health_Lifestyle", print_info = False, export_csv = False, drop_cols = ['id'], Standard =False):
 
     data_type = "binary"
-    csv_path = f'./exp_data/{data_name}/health_lifestyle_dataset.csv'
+    csv_path = f'exp_data/{data_name}/health_lifestyle_dataset.csv'
     
     label_col = 'disease_risk'
-    label_map = {0: -1, 1: 1}
+    label_map = {0: 0, 1: 1}
     
 
     df = _run(csv_path, data_name, data_type, drop_cols, label_col, label_map, print_info, export_csv=export_csv)
@@ -215,7 +215,7 @@ def health_lifestyle(data_name = "Health Lifestyle", print_info = False, export_
     return df
 
 
-def medical_insurance_cost_prediction(data_name = "Medical Insurance Cost Prediction", print_info = False, export_csv = False, drop_cols = ['alcohol_freq'], Standard = False):
+def medical_insurance_cost_prediction(data_name = "Medical_Insurance_Cost Prediction", print_info = False, export_csv = False, drop_cols = ['alcohol_freq'], Standard = False):
     """
     1. The missing values in this dataset are handled by directly removing the corresponding column. Since the `alcohol_freq` column contains a large number of missing values, deleting the rows would result in significant data loss, so the entire column is dropped instead.
 
@@ -223,7 +223,7 @@ def medical_insurance_cost_prediction(data_name = "Medical Insurance Cost Predic
     """
 
     data_type = "binary"
-    csv_path = f'./exp_data/{data_name}/medical_insurance.csv'
+    csv_path = f'exp_data/{data_name}/medical_insurance.csv'
     
     label_col = 'is_high_risk'
     label_map = {0: -1, 1: 1}
@@ -234,10 +234,10 @@ def medical_insurance_cost_prediction(data_name = "Medical Insurance Cost Predic
     return df
 
 
-def particle_physics_event_classification(data_name = "Particle Physics Event Classification", print_info = False, export_csv = False, drop_cols = [], Standard =False):
+def particle_physics_event_classification(data_name = "Particle_Physics_Event_Classification", print_info = False, export_csv = False, drop_cols = [], Standard =False):
 
     data_type = "binary"
-    csv_path = f'./exp_data/{data_name}/Particle Physics Event Classification.csv'
+    csv_path = f'exp_data/{data_name}/Particle Physics Event Classification.csv'
     
     label_col = 'Label'
     label_map = {'s': -1, 'b': 1}
@@ -249,13 +249,13 @@ def particle_physics_event_classification(data_name = "Particle Physics Event Cl
 
 
 
-def adult_income_prediction(data_name = "Adult Income Prediction", print_info = False, export_csv=False, drop_cols = [], Standard = False):
+def adult_income_prediction(data_name = "Adult_Income_Prediction", print_info = False, export_csv=False, drop_cols = [], Standard = False):
 
     data_type = "binary"
     csv_path = f'./exp_data/{data_name}/adult.csv'
     
     label_col = 'income'
-    label_map = {'<=50K': -1, '>50K': 1}
+    label_map = {'<=50K': 0, '>50K': 1}
     
 
     df = _run(csv_path, data_name, data_type, drop_cols, label_col, label_map, print_info, export_csv=export_csv)
@@ -263,13 +263,13 @@ def adult_income_prediction(data_name = "Adult Income Prediction", print_info =
     return df
 
 
-def TamilNadu_weather_2020_2025(data_name = "TN Weather 2020-2025", print_info = False, export_csv = False, drop_cols = ['Unnamed: 0'], Standard = False):
+def TamilNadu_weather_2020_2025(data_name = "TN_Weather_2020_2025", print_info = False, export_csv = False, drop_cols = ['Unnamed: 0'], Standard = False):
 
     data_type = "binary"
     csv_path = f'./exp_data/{data_name}/TNweather_1.8M.csv'
     
     label_col = 'rain_tomorrow'
-    label_map = {0: -1, 1: 1}
+    label_map = {0: 0, 1: 1}
 
     time_info = {
         'time_col_name': 'time',
@@ -281,7 +281,7 @@ def TamilNadu_weather_2020_2025(data_name = "TN Weather 2020-2025", print_info =
 
     return df
 
-def YouTube_Recommendation(data_name = "YouTube Recommendation", print_info = False, export_csv = False, drop_cols = ['user_id']):
+def YouTube_Recommendation(data_name = "YouTube_Recommendation", print_info = False, export_csv = False, drop_cols = ['user_id']):
 
     data_type = "binary"
     csv_path = f'./exp_data/{data_name}/youtube recommendation dataset.csv'
@@ -303,13 +303,13 @@ def YouTube_Recommendation(data_name = "YouTube Recommendation", print_info = Fa
     return df
 
 
-def Santander_Customer_Satisfaction(data_name = "SantanderCustomerSatisfaction", print_info = False, export_csv = False):
+def Santander_Customer_Satisfaction(data_name = "Santander_Customer_Satisfaction", print_info = False, export_csv = False):
     data_type = "binary"
     csv_path = None
 
     drop_cols = ['ID_code']
     label_col = 'target'
-    label_map = {False: -1, True: 1}
+    label_map = {False: 0, True: 1}
 
     df, y, categorical_indicator, attribute_names = junshan_kit.kit.download_openml_data(data_name)
 
@@ -324,7 +324,7 @@ def newsgroups_drift(data_name = "20_newsgroups.drift", print_info = False, expo
 
     drop_cols = ['ID_code']
     label_col = 'target'
-    label_map = {False: -1, True: 1}
+    label_map = {False: 0, True: 1}
 
     df, y, categorical_indicator, attribute_names = junshan_kit.kit.download_openml_data(data_name)
 
@@ -340,7 +340,7 @@ def Homesite_Quote_Conversion(data_name = "Homesite_Quote_Conversion", print_inf
 
     drop_cols = ['QuoteNumber']
     label_col = 'QuoteConversion_Flag'
-    label_map = {0: -1, 1: 1}
+    label_map = {0: 0, 1: 1}
 
     time_info = {
         'time_col_name': 'Original_Quote_Date',
@@ -353,7 +353,6 @@ def Homesite_Quote_Conversion(data_name = "Homesite_Quote_Conversion", print_inf
 
     return df
 
-
 def IEEE_CIS_Fraud_Detection(data_name = "IEEE-CIS_Fraud_Detection", print_info = False, export_csv = False, export_mat = False):
     data_type = "binary"
     csv_path = None
@@ -361,7 +360,7 @@ def IEEE_CIS_Fraud_Detection(data_name = "IEEE-CIS_Fraud_Detection", print_info
 
     drop_cols = ['TransactionID']
     label_col = 'isFraud'
-    label_map = {0: -1, 1: 1}
+    label_map = {0: 0, 1: 1}
 
     Paras = {
         "export_mat": export_mat

junshan_kit/Evaluate_Metrics.py

@@ -2,7 +2,7 @@ import torch
 from torch.nn.utils import parameters_to_vector
 import torch.nn.functional as F
 
-def compute_epoch_loss(X, y, model, loss_fn, Paras):
+def loss(X, y, model, loss_fn, Paras):
     pred = model(X)
     _, c = pred.shape
 
@@ -37,4 +37,77 @@ def compute_epoch_loss(X, y, model, loss_fn, Paras):
             )
             assert False
 
-    return loss
+    return loss
+
+def compute_loss_acc(X, y, model, loss_fn, Paras):
+    pred = model(X)
+    m, c = pred.shape
+
+    if c == 1:
+        # Logistic Regression (binary)
+        if isinstance(loss_fn, torch.nn.BCEWithLogitsLoss):
+            pred = pred.view(-1).float()
+            loss = loss_fn(pred, y).item()
+
+            if Paras["model_name"] == "LogRegressionBinaryL2":
+                x = parameters_to_vector(model.parameters())
+                lam = Paras["lambda"]
+                loss = (loss + 0.5 * lam * torch.norm(x, p=2) ** 2).item()
+
+            pred_label = (torch.sigmoid(pred) > 0.5).float()
+            correct = (pred_label == y).sum().item()
+
+        else:
+            assert False
+
+    else:
+
+        # Least Square （mutil）
+        if isinstance(loss_fn, torch.nn.MSELoss):
+            # loss
+            y_onehot = F.one_hot(y.long(), num_classes=c).float()
+            pred_label = pred.argmax(1).long()
+            pred_ont = F.one_hot(pred_label, num_classes=c).float()
+            loss = 0.5 * loss_fn(pred_ont, y_onehot).item() * c
+
+            # acc
+            correct = (pred_label == y).sum().item()
+
+        elif isinstance(loss_fn, torch.nn.CrossEntropyLoss):
+
+            # loss
+            loss = loss_fn(pred, y.long()).item()
+
+            # acc
+            # acc
+            pred_label = pred.argmax(1).long()
+            correct = (pred_label == y).sum().item()
+
+        else:
+            print(
+                f"\033[34m **** isinstance(loss_fn, torch.nn.MSELoss)? {isinstance(loss_fn, torch.nn.MSELoss)} **** \033[0m"
+            )
+            assert False
+
+    return loss, correct
+
+def get_loss_acc(dataloader, model, loss_fn, Paras):
+    # model.eval()
+    size = len(dataloader.dataset)
+    num_batches = len(dataloader)
+    loss, correct = 0, 0
+
+    device = Paras["device"]
+
+    with torch.no_grad():
+        for X, y in dataloader:
+            X, y = X.to(device).float(), y.to(device).float()
+            per_loss, per_acc = compute_loss_acc(X, y, model, loss_fn, Paras)
+
+            loss += per_loss
+            correct += per_acc
+
+    loss /= num_batches
+    correct /= size
+
+    return loss, correct