Adjust default values and references

JesusEV · Sep 17, 2024 · 51fec15 · 51fec15
1 parent 5b31105
commit 51fec15
Show file tree

Hide file tree

Showing 4 changed files with 28 additions and 25 deletions.
diff --git a/pynest/examples/eprop_plasticity/eprop_supervised_classification_evidence-accumulation.py b/pynest/examples/eprop_plasticity/eprop_supervised_classification_evidence-accumulation.py
@@ -121,8 +121,8 @@
 parser.add_argument("--recordings_dir", type=str, default="./")
 parser.add_argument("--seed", type=int, default=1)
 parser.add_argument("--surrogate_gradient", type=str.lower, default="piecewise_linear")
-parser.add_argument("--surrogate_gradient_beta", type=float, default=1.0)
-parser.add_argument("--surrogate_gradient_gamma", type=float, default=0.3)
+parser.add_argument("--surrogate_gradient_beta", type=float, default=1.7)
+parser.add_argument("--surrogate_gradient_gamma", type=float, default=0.5)
 
 args = parser.parse_args()
 

diff --git a/pynest/examples/eprop_plasticity/eprop_supervised_classification_neuromorphic_mnist.py b/pynest/examples/eprop_plasticity/eprop_supervised_classification_neuromorphic_mnist.py
@@ -121,8 +121,8 @@
 parser.add_argument("--recordings_dir", type=str, default="./")
 parser.add_argument("--seed", type=int, default=1)
 parser.add_argument("--surrogate_gradient", type=str.lower, default="piecewise_linear")
-parser.add_argument("--surrogate_gradient_beta", type=float, default=33.3)
-parser.add_argument("--surrogate_gradient_gamma", type=float, default=10.0)
+parser.add_argument("--surrogate_gradient_beta", type=float, default=1.7)
+parser.add_argument("--surrogate_gradient_gamma", type=float, default=0.5)
 
 args = parser.parse_args()
 

diff --git a/...mples/eprop_plasticity/eprop_supervised_classification_neuromorphic_mnist_bsshslm_2020.py b/...mples/eprop_plasticity/eprop_supervised_classification_neuromorphic_mnist_bsshslm_2020.py
@@ -242,6 +242,9 @@
     "V_th": 0.6,  # mV, spike threshold membrane voltage
 }
 
+params_nrn_rec["gamma"] /= params_nrn_rec["V_th"]
+params_nrn_rec["beta"] /= np.abs(params_nrn_rec["V_th"])
+
 ####################
 
 # Intermediate parrot neurons required between input spike generators and recurrent neurons,
@@ -613,7 +616,7 @@ def create_input_output(loader, t_start_iteration, t_end_iteration, target_signa
     return params_gen_spk_in, params_gen_rate_target
 
 
-save_path = "./"  # path to save the N-MNIST dataset to
+save_path = "./"  # path to save the N-MNIST dataset to'
 train_path, test_path = download_and_extract_nmnist_dataset(save_path)
 
 selected_labels = [label for label in range(n_out)]

diff --git a/pynest/examples/eprop_plasticity/toolbox.py b/pynest/examples/eprop_plasticity/toolbox.py
@@ -199,11 +199,11 @@ def verify(self):
             ]
         elif file_name == "eprop_supervised_classification_evidence-accumulation.py":
             loss_reference = [
-                33.99707684974502,
-                34.54963564066860,
-                36.15785645581101,
-                35.49256076087588,
-                35.86926256681858,
+                34.90337404416763,
+                34.78511921272757,
+                35.18474914160545,
+                35.33582702752958,
+                35.22131262247200,
             ]
         elif file_name == "eprop_supervised_regression_sine-waves_bsshslm_2020.py":
             loss_reference = [
@@ -215,27 +215,27 @@ def verify(self):
             ]
         elif file_name == "eprop_supervised_regression_sine-waves.py":
             loss_reference = [
-                102.32452148947576,
-                102.02905791284471,
-                104.18266745765692,
-                104.67929923397385,
-                105.13095153301973,
+                102.32452148947574,
+                103.06940389745640,
+                102.58440415993280,
+                103.95247660721655,
+                105.23404961833215,
             ]
         elif file_name == "eprop_supervised_classification_neuromorphic_mnist.py":
             loss_reference = [
-                0.49165621803559,
-                0.50931366438794,
-                0.50123237065641,
-                0.49169665763603,
-                0.45463618073378,
+                0.50059986996625,
+                0.50130915105991,
+                0.49801405188182,
+                0.49384438941558,
+                0.47842946665255,
             ]
         elif file_name == "eprop_supervised_classification_neuromorphic_mnist_bsshslm_2020.py":
             loss_reference = [
-                2.29905872344452,
-                2.31039108412163,
-                2.29426963047102,
-                2.29407265212433,
-                2.29758026605794,
+                2.29923887455666,
+                2.30994402967322,
+                2.29404627571308,
+                2.29464031430063,
+                2.29825169987960,
             ]
         elif file_name == "eprop_supervised_regression_lemniscate_bsshslm_2020.py":
             loss_reference = [