cleanup(json generation)

ref_cnn/generate_cnn_json.py

@@ -155,10 +155,6 @@ class Layer(Enum):
     "data": initial
     }
 
-json_data = json.dumps(data, cls=Encoder)
-
-layers_json = {"initial":  json_data}
-
 # conv layer
 shape = (32,5,5,3)
 f = np.random.randint(low=-10, high=+10, size=shape) 
@@ -178,9 +174,6 @@ class Layer(Enum):
 }
 
 model = [conv]
-json_data = json.dumps(data, cls=Encoder)
-
-layers_json["conv1"] = json_data
 
 x, n_params, n_multiplications, name = conv_layer(x, f)
 print(p.format(name, str(x.shape), n_params, n_multiplications))
@@ -229,10 +222,6 @@ class Layer(Enum):
 
 model.append(conv)
 
-json_data = json.dumps(data, cls=Encoder)
-
-layers_json["conv2"] = json_data
-
 x, n_params, n_multiplications, name = conv_layer(x, f)
 print(p.format(name, str(x.shape), n_params, n_multiplications))
 
@@ -272,31 +261,24 @@ class Layer(Enum):
 
 # fully connected
 shape = (1000, x.shape[0])
-weights = np.random.randint(low=-10, high=+10, size=shape) 
+weights = np.random.randint(low=-10, high=+10, size=shape)
 
 # weights json
-weights1 = weights.flatten().astype(np.float32, copy=False)
-
 data = {
     "v": 1,
     "dim": shape,
-    "data": weights1
+    "data": weights.flatten().astype(np.float32, copy=False)
 }
 
-json_data = json.dumps(data, cls=Encoder)
-
-layers_json["weights1"] = json_data
-
 # biases json
 shape = [1000]
 biases = np.random.randint(low=-10, high=+10, size=(1000))
-biases1 = biases.flatten().astype(np.float32, copy=False)
 
 data2 = {
     "v": 1,
     # ndarray can't take a single value, needs to be in json array
     "dim": shape,
-    "data": biases1
+    "data": biases.flatten().astype(np.float32, copy=False)
 }
 
 fully_connected = {
@@ -308,10 +290,6 @@ class Layer(Enum):
 
 model.append(fully_connected)
 
-json_data = json.dumps(data2, cls=Encoder)
-
-layers_json["biases1"] = json_data
-
 x, n_params, n_multiplications, name = fully_connected_layer(x, weights, biases)
 print(p.format(name, str(x.shape), n_params, n_multiplications))
 
@@ -328,32 +306,22 @@ class Layer(Enum):
 
 # fully connected
 shape = (5, x.shape[0])
-weights = np.random.randint(low=-10, high=+10, size=shape) 
-
-weights2 = weights.flatten().astype(np.float32, copy=False)
+weights = np.random.randint(low=-10, high=+10, size=shape)
 
 data = {
     "v": 1,
     "dim": shape,
-    "data": weights2
+    "data": weights.flatten().astype(np.float32, copy=False)
 }
 
-json_data = json.dumps(data, cls=Encoder)
-
-layers_json["weights2"] = json_data
-
-
 shape = [5]
-biases = np.random.randint(low=-10, high=+10, size=shape) 
-
-
-biases2 = biases.flatten().astype(np.float32, copy=False)
+biases = np.random.randint(low=-10, high=+10, size=shape)
 
 data2 = {
     "v": 1,
     # ndarray can't take a single value, needs to be in json array
     "dim": [5],
-    "data": biases2
+    "data": biases.flatten().astype(np.float32, copy=False)
 }
 
 fully_connected = {
@@ -365,17 +333,6 @@ class Layer(Enum):
 
 model.append(fully_connected)
 
-json_data = json.dumps(data2, cls=Encoder)
-
-layers_json["biases2"] = json_data
-
-# create files for arrays
-for layer, json_data in layers_json.items():
-    with open(f'../src/json/{layer}.json', "w") as f:
-        print(f'created {layer}.json in the proto-neural-zkp/src/json folder')
-        f.write(json_data)
-
-
 x, n_params, n_multiplications, name = fully_connected_layer(x, weights, biases)
 print(p.format(name, str(x.shape), n_params, n_multiplications))
 
@@ -396,10 +353,11 @@ class Layer(Enum):
 x, n_params, n_multiplications, name = normalize(x)
 print(p.format(name, str(x.shape), n_params, n_multiplications))
 
-model_data = json.dumps(model, cls=Encoder)
-with open('../src/json/model.json', "w") as f:
-    print('created model.json in the proto-neural-zkp/src/json folder')
-    f.write(model_data)
 
+print("\nfinal output:", x)
 
-print("\nfinal output:", x)
+
+model_data = json.dumps(model, cls=Encoder)
+with open('../src/json/model.json', "w") as f:
+    print('\ncreated model.json in the proto-neural-zkp/src/json folder')
+    f.write(model_data)

ref_cnn/vanilla_cnn.ipynb

@@ -226,6 +226,17 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "import json\n",
+    "\n",
+    "# Encoder\n",
+    "from json import JSONEncoder\n",
+    "\n",
+    "class Encoder(JSONEncoder):\n",
+    "    def default(self, obj):\n",
+    "        if isinstance(obj, np.ndarray):\n",
+    "            return obj.tolist()\n",
+    "        return JSONEncoder.default(self, obj)\n",
+    "\n",
     "np.random.seed(12345)\n",
     "\n",
     "shape = (3,3,3)\n",
@@ -240,7 +251,7 @@
     "    \"data\": x\n",
     "    }\n",
     "\n",
-    "json_data = json.dumps(data, cls=NumpyArrayEncoder)\n",
+    "json_data = json.dumps(data, cls=Encoder)\n",
     "\n",
     "with open(\"../src/json/test.json\", \"w\") as f:\n",
     "    f.write(json_data)"