improve lbfgs in hotrg, but still performs worse then hotrg_sweep

hotrg.py

@@ -58,8 +58,8 @@ if __name__=='__main__':
     parser = argparse.ArgumentParser(description='')
     
     parser.add_argument("-beta", type=float, default=0.44, help="beta")
-    parser.add_argument("-chi", type=int, default=20, help="chi")
-    parser.add_argument("-niter", type=int, default=20, help="niter")
+    parser.add_argument("-chi", type=int, default=24, help="chi")
+    parser.add_argument("-niter", type=int, default=30, help="niter")
     parser.add_argument("-Nepochs", type=int, default=100, help="Nepochs")
     parser.add_argument("-float32", action='store_true', help="use float32")
     parser.add_argument("-cuda", type=int, default=-1, help="GPU #")
@@ -71,7 +71,8 @@ if __name__=='__main__':
 
     model = Ising(args.chi, args.niter, dtype=dtype, device=device)
 
-    optimizer = torch.optim.LBFGS(model.parameters(), max_iter=10)
+    optimizer = torch.optim.LBFGS(model.parameters(), max_iter=10, tolerance_grad=0, tolerance_change=0, line_search_fn='strong_wolfe')
+
     params = list(model.parameters())
     params = list(filter(lambda p: p.requires_grad, params))
     nparams = sum([np.prod(p.size()) for p in params])
@@ -85,4 +86,4 @@ if __name__=='__main__':
 
     for epoch in range(args.Nepochs):
         loss = optimizer.step(closure)
-        print (epoch, loss.item())
+        print (epoch, loss.item(), model.A.grad.abs().max().item())