feat(baselines) editing simII data generation

baselines/fedht/fedht/client.py

@@ -88,8 +88,8 @@ def client_fn(context: Context) -> Client:
         test_dataset = train_dataset = MyDataset(
             X_train[int(partition_id), :, :], y_train[:, int(partition_id)]
         )
-        trainloader = DataLoader(train_dataset, batch_size=cfg.batch_size, shuffle=True)
-        testloader = DataLoader(test_dataset, batch_size=cfg.batch_size, shuffle=True)
+        trainloader = DataLoader(train_dataset, batch_size=cfg.batch_size, shuffle=False)
+        testloader = DataLoader(test_dataset, batch_size=cfg.batch_size, shuffle=False)
 
         # define model and set device
         device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

baselines/fedht/fedht/main.py

@@ -3,6 +3,7 @@
 import pickle
 import random
 import torch
+import gzip
 
 import flwr as fl
 import hydra
@@ -68,23 +69,21 @@ def main(cfg: DictConfig):
         num_classes = cfg.num_classes
 
         # import data from fedht/data folder
-        with open('fedht/data/simII_train.pkl', 'rb') as file:
-            dataset = pickle.load(file)
+        # with open('fedht/data/simII_train.pkl', 'rb') as file:
+        #     dataset = pickle.load(file)
 
-        with open('fedht/data/simII_test.pkl', 'rb') as file:
-            test_dataset = pickle.load(file)
+        # with open('fedht/data/simII_test.pkl', 'rb') as file:
+        #     test_dataset = pickle.load(file)
 
-        X_test, y_test = test_dataset
-
-        # dataset = sim_data(num_obs, num_clients, num_features, 1, 1)
-        # X_test, y_test = sim_data(num_obs, 1, num_features, 1, 1)
-        # test_dataset = MyDataset(X_test[0, :, :], y_test[:, 0])
+        # simulate data
+        X_train, y_train, X_test, y_test = sim_data(200, num_clients, 1000, 1, 1)
+        train_dataset = X_train, y_train
         test_dataset = MyDataset(X_test, y_test[:,0])
         testloader = DataLoader(test_dataset, batch_size=cfg.batch_size, shuffle=False)
 
         # set client function
         client_fn = generate_client_fn_simII(
-            dataset,
+            train_dataset,
             cfg=cfg
         )
 

baselines/fedht/fedht/utils.py

@@ -49,12 +49,12 @@ def partition_data(data, num_partitions):
 def sim_data(ni: int, num_clients: int, num_features: int, alpha=1, beta=1):
     """Simulate data for simII."""
 
-    np.random.seed(2025)
+    np.random.seed(2024)
 
     # generate client-based model coefs
     u = np.random.normal(0, alpha, num_clients)
     x = np.zeros((num_features, num_clients))
-    x[0:99, :] = np.random.multivariate_normal(u, np.diag(np.ones(num_clients)), 99)
+    x[0:100, :] = np.random.multivariate_normal(u, np.diag(np.ones(num_clients)), 100)
 
     # generate train observations
     ivec = np.arange(1, num_features + 1)
@@ -89,15 +89,17 @@ def sim_data(ni: int, num_clients: int, num_features: int, alpha=1, beta=1):
     # B = np.random.normal(0, beta, num_features)
     # v = np.random.multivariate_normal(B, np.diag(np.ones(num_features)), num_clients)
 
-    error = np.random.multivariate_normal(u, np.diag(np.ones(num_clients)), ntest)
+    error2 = np.random.multivariate_normal(u, np.diag(np.ones(num_clients)), ntest)
+    xtest = np.zeros((num_features, num_clients))
+    xtest[0:100, :] = np.random.multivariate_normal(u, np.diag(np.ones(num_clients)), 100)
     ztest = np.zeros((num_clients, ntest, num_features))
     ytest = np.zeros((ni, num_clients))
 
     # (num_clients, ni, num_feaures)
     for i in range(ztest.shape[0]):
         # train
         ztest[i, :, :] = np.random.multivariate_normal(v[i], vari, ntest)
-        hold = np.matmul(ztest[i, :, :], x[:, i]) + error[:, i]
+        hold = np.matmul(ztest[i, :, :], xtest[:, i]) + error2[:, i]
         ytest[:, i] = np.exp(hold) / (1 + np.exp(hold))
 
     for j in range(num_clients):