a few updates on risk curve

lilianweng · Mar 21, 2019 · 7593d04 · 7593d04
1 parent c2293e6
commit 7593d04
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 6 deletions.
diff --git a/risk_curve.py b/risk_curve.py
@@ -24,7 +24,7 @@ def __init__(self, loss_type, max_epochs, n_train_sample):
         logging.info(f"critical_n_units: {critical_n_units}")
         self.n_units_to_test = sorted(set(
             list(range(critical_n_units - 7, critical_n_units + 4)) +
-            list(range(5, 105, 10))
+            list(range(5, 55, 5)) + list(range(50, 105, 10)) + [120, 150, 200]
         ))
         logging.info(f"n_units_to_test: {self.n_units_to_test}")
 
@@ -43,7 +43,7 @@ def run(self):
                 '--loss-type', str(self.loss_type),
             ]
 
-            if old_n_units and total_params < self.n_train_sample * 10:
+            if old_n_units:  # and total_params < self.n_train_sample * 10:
                 args.extend(['--old-n-units', str(old_n_units)])
 
             proc = subprocess.run(
@@ -78,5 +78,5 @@ def plot(self):
 
 
 if __name__ == '__main__':
-    exp = NewRiskCurveExperiment(loss_type='mse', max_epochs=500, n_train_sample=2500)
+    exp = NewRiskCurveExperiment(loss_type='mse', max_epochs=500, n_train_sample=4000)
     exp.run()
diff --git a/risk_curve_evaluate_model.py b/risk_curve_evaluate_model.py
@@ -108,12 +108,12 @@ def report_performance(sess, n_units, old_n_units, max_epochs, loss_type, lr, ba
 @click.option('--old-n-units', default=None, type=int, help="")
 @click.option('--loss-type', default='mse', type=str, help="type of loss func.")
 @click.option('--max-epochs', default=500, type=int, help="num. training epochs.")
-@click.option('--n-train-samples', default=2500, type=int, help="num. training samples")
-def main(n_units=1, old_n_units=None, loss_type='mse', max_epochs=500, n_train_samples=2500):
+@click.option('--n-train-samples', default=4000, type=int, help="num. training samples")
+def main(n_units=1, old_n_units=None, loss_type='mse', max_epochs=500, n_train_samples=4000):
     assert old_n_units is None or old_n_units < n_units
     logging.info(f"n_units:{n_units} max_epochs:{max_epochs}")
     sess = make_session()
-    lr = 0.005
+    lr = 0.001
     batch_size = 128
 
     epoch, step, train_loss, train_acc, eval_loss, eval_acc = report_performance(