change hyperparameters.json to hyperparameters.yaml

README.md

@@ -129,7 +129,7 @@ notebook for an overview of the Python API, including fitting your own predictor
 An ensemble of eight single-allele models was trained for each allele with at least
 100 measurements in the training set (118 alleles). The models were trained on a
 random 80% sample of the data for the allele and the remaining 20% was used for
-early stopping. All models use the same [architecture](downloads-generation/models_class1/hyperparameters.json). The
+early stopping. All models use the same [architecture](downloads-generation/models_class1/hyperparameters.yaml). The
 predictions are taken to be the geometric mean of the nM binding affinity
 predictions of the individual models. The training script is [here](downloads-generation/models_class1/GENERATE.sh).
 

downloads-generation/models_class1/GENERATE.sh

@@ -25,11 +25,11 @@ cd $SCRATCH_DIR/$DOWNLOAD_NAME
 
 mkdir models
 
-cp $SCRIPT_DIR/hyperparameters.json .
+cp $SCRIPT_DIR/hyperparameters.yaml .
 
 time mhcflurry-class1-train-allele-specific-models \
     --data "$(mhcflurry-downloads path data_curated)/curated_training_data.csv.bz2" \
-    --hyperparameters hyperparameters.json \
+    --hyperparameters hyperparameters.yaml \
     --out-models-dir models \
     --min-measurements-per-allele 200
 

downloads-generation/models_class1/hyperparameters.json

-[
-    {
-        "n_models": 12,
-        "max_epochs": 500,
-        "patience": 10,
-        "early_stopping": true,
-        "validation_split": 0.2,
-
-        "random_negative_rate": 0.0,
-        "random_negative_constant": 25,
-
-        "use_embedding": false,
-        "kmer_size": 15,
-        "batch_normalization": false,
-        "locally_connected_layers": [
-            {
-                "filters": 8,
-                "activation": "tanh",
-                "kernel_size": 3
-            },
-            {
-                "filters": 8,
-                "activation": "tanh",
-                "kernel_size": 3
-            }
-        ],
-        "activation": "relu",
-        "output_activation": "sigmoid",
-        "layer_sizes": [
-            32
-        ],
-        "random_negative_affinity_min": 20000.0,
-        "random_negative_affinity_max": 50000.0,
-        "dense_layer_l1_regularization": 0.001,
-        "dropout_probability": 0.0
-    }
-]

downloads-generation/models_class1/hyperparameters.yaml

+[{
+##########################################
+# ENSEMBLE SIZE
+##########################################
+"n_models": 12,
+
+##########################################
+# OPTIMIZATION
+##########################################
+"max_epochs": 500,
+"patience": 10,
+"early_stopping": true,
+"validation_split": 0.2,
+
+##########################################
+# RANDOM NEGATIVE PEPTIDES
+##########################################
+"random_negative_rate": 0.0,
+"random_negative_constant": 25,
+"random_negative_affinity_min": 20000.0,
+"random_negative_affinity_max": 50000.0,
+
+##########################################
+# PEPTIDE REPRESENTATION
+##########################################
+# One of "one-hot", "embedding", or "BLOSUM62".
+"peptide_amino_acid_encoding": "BLOSUM62",
+"use_embedding": false,  # maintained for backward compatability
+"kmer_size": 15,
+
+##########################################
+# NEURAL NETWORK ARCHITECTURE
+##########################################
+"locally_connected_layers": [
+    {
+        "filters": 8,
+        "activation": "tanh",
+        "kernel_size": 3
+    },
+    {
+        "filters": 8,
+        "activation": "tanh",
+        "kernel_size": 3
+    }
+],
+"activation": "relu",
+"output_activation": "sigmoid",
+"layer_sizes": [
+    32
+],
+"dense_layer_l1_regularization": 0.001,
+"batch_normalization": false,
+"dropout_probability": 0.0,
+}]