diff --git a/src/convnet_magicnet.js b/src/convnet_magicnet.js
index 7f3c159c..a60d536b 100644
--- a/src/convnet_magicnet.js
+++ b/src/convnet_magicnet.js
@@ -50,6 +50,10 @@
     this.momentum_max = getopt(opt, 'momentum_max', 0.9);
     this.neurons_min = getopt(opt, 'neurons_min', 5);
     this.neurons_max = getopt(opt, 'neurons_max', 30);
+    
+    this.input_out_sx = getopt(opt, 'input_out_sx', 1);
+    this.input_out_sy = getopt(opt, 'input_out_sy', 1);
+    this.input_out_depth = getopt(opt, 'input_out_depth', this.data[0].w.length);
 
     // computed
     this.folds = []; // data fold indices, gets filled by sampleFolds()
@@ -85,12 +89,11 @@
 
     // returns a random candidate network
     sampleCandidate: function() {
-      var input_depth = this.data[0].w.length;
       var num_classes = this.unique_labels.length;
 
       // sample network topology and hyperparameters
       var layer_defs = [];
-      layer_defs.push({type:'input', out_sx:1, out_sy:1, out_depth: input_depth});
+      layer_defs.push({type: 'input', out_sx: this.input_out_sx, out_sy: this.input_out_sy, out_depth: this.input_out_depth});
       var nl = weightedSample([0,1,2,3], [0.2, 0.3, 0.3, 0.2]); // prefer nets with 1,2 hidden layers
       for(var q=0;q<nl;q++) {
         var ni = randi(this.neurons_min, this.neurons_max);