Fixes calculuation of hat values

utils/sherlock_script.py

@@ -12,9 +12,14 @@
 import torch
 
 import sys
-if len(sys.argv) != 2:
+if len(sys.argv) == 1 or len(sys.argv) > 3:
     raise RuntimeError()
 
+elif len(sys.argv) == 2:
+    task_id = 0xEE364A
+else len(sys.argv) == 3:
+    task_id = int(sys.argv[2])
+
 data_dir = "/oak/stanford/groups/candes/for_parth"
 cache_dir = "/scratch/groups/candes/parth"
 df = pd.read_csv(os.path.join(data_dir, "phenotypes.QC.britishonly.csv"), index_col=0)
@@ -38,7 +43,7 @@
 )
 print(X.shape)
 
-rng = np.random.default_rng(0xEE364A)
+rng = np.random.default_rng(task_id)
 P = np.random.permutation(y.shape[-1])
 n_train = P.size * 9 // 10
 train_mask = P[:n_train]
@@ -61,12 +66,12 @@
 L = state.betas.shape[0]
 oos = np.empty(L)
 ins = np.empty(L)
-y_hat_test = X_test @ state.betas.T
-y_hat_train = X_train @ state.betas.T
+y_hat_test = ad.diagnostic.predict(X_test, state.betas, state.intercepts)
+y_hat_train = ad.diagnostic.predict(X_train, state.betas, state.intercepts)
 for i in range(L):
-    oos[i] = loss(torch.from_numpy(y_hat_test[:, i]), torch.from_numpy(y_test))
-    ins[i] = loss(torch.from_numpy(y_hat_train[:, i]), torch.from_numpy(y_train))
+    oos[i] = loss(torch.from_numpy(y_hat_test[i]), torch.from_numpy(y_test))
+    ins[i] = loss(torch.from_numpy(y_hat_train[i]), torch.from_numpy(y_train))
 
 ld, alo, ts, r2 = ai.get_alo_for_sweep(y_train, state, loss)
 
-np.savez(sys.argv[-1], lamda=ld, alo=alo, oos=oos, in_sample=ins, ts=ts, r2=r2)
+np.savez(sys.argv[1], lamda=ld, alo=alo, oos=oos, in_sample=ins, ts=ts, r2=r2)