diff --git a/downloads-generation/models_class1_pan/generate_hyperparameters.py b/downloads-generation/models_class1_pan/generate_hyperparameters.py
index fe05e31354d3cdc06518f17486eb65187fcc3ffd..ea728dc31aea8bc6660da38164dd5499a93445bb 100644
--- a/downloads-generation/models_class1_pan/generate_hyperparameters.py
+++ b/downloads-generation/models_class1_pan/generate_hyperparameters.py
@@ -48,7 +48,7 @@ base_hyperparameters = {
 }
 
 grid = []
-for layer_sizes in [[1024, 512], [512, 512], [1024, 1024]]:
+for layer_sizes in [[512, 256], [1024, 512], [1024, 1024]]:
     for l1 in [0.0, 0.0001, 0.001, 0.01]:
         new = deepcopy(base_hyperparameters)
         new["layer_sizes"] = layer_sizes
diff --git a/mhcflurry/train_pan_allele_models_command.py b/mhcflurry/train_pan_allele_models_command.py
index 9880f804373e1ea9d0bb46a8a5b796439b24b58d..fa530d152d80ed09ee66bd7e1d749d33c5931540 100644
--- a/mhcflurry/train_pan_allele_models_command.py
+++ b/mhcflurry/train_pan_allele_models_command.py
@@ -434,7 +434,7 @@ def train_model(
     train_peptides = EncodableSequences(train_data.peptide.values)
     train_alleles = AlleleEncoding(
         train_data.allele.values, borrow_from=allele_encoding)
-    train_target = from_ic50(train_data.measurement_value)
+    train_target = from_ic50(train_data.measurement_value.values)
 
     model = Class1NeuralNetwork(**hyperparameters)
 
@@ -468,6 +468,7 @@ def train_model(
                 peptides=peptides,
                 affinities=affinities,
                 allele_encoding=alleles)
+
             fit_time = time.time() - start
             start = time.time()
             predictions = model.predict(
@@ -484,7 +485,7 @@ def train_model(
                 mask = train_data.measurement_inequality == inequality
                 predictions[mask.values] = func(
                     predictions[mask.values],
-                    train_data.loc[mask.values].measurement_value.values)
+                    train_data.loc[mask].measurement_value.values)
             score_mse = numpy.mean((from_ic50(predictions) - train_target)**2)
             score_time = time.time() - start
             print(