Merge NMED functionality.

cme433_lrtestbench-main/python/get_mnist_stats.py

@@ -2,6 +2,14 @@ import gzip
 import sys
 import numpy as np
 
+SVERILOG_BATCH_COUNT = 100
+SVERILOG_BATCH_SIZE = 1
+ROOT_DIR = "../results/"
+SVERILOG_FIRST_LAYER = 0
+SVERILOG_SECOND_LAYER = 1
+SVERILOG_FINAL_LAYER = 2
+MAX_VALUE = 16129 # 127 * 127
+VERSION = sys.argv[1]
 
 def twoscomp_to_decimal(inarray, bits_in_word):
     inputs = []
@@ -13,7 +21,6 @@ def twoscomp_to_decimal(inarray, bits_in_word):
     inputs = np.array(inputs)
     return inputs
 
-
 test_y = None
 with open("../data/t10k-labels-idx1-ubyte.gz", "rb") as f:
     data = f.read()
@@ -21,14 +28,6 @@ with open("../data/t10k-labels-idx1-ubyte.gz", "rb") as f:
 
 test_y = test_y[8:]
 
-# print(sys.argv)
-
-SVERILOG_BATCH_COUNT = 100
-SVERILOG_BATCH_SIZE = 1
-ROOT_DIR = "../results/"
-SVERILOG_FINAL_LAYER = 2
-VERSION = sys.argv[1]
-
 acc = 0
 for i in range(0, SVERILOG_BATCH_COUNT):
     predictions_bin = []
@@ -44,3 +43,61 @@ for i in range(0, SVERILOG_BATCH_COUNT):
         acc += 1
 
 print("Acc: ", acc * 100 / SVERILOG_BATCH_COUNT)
+
+"""
+    Calculate the Normalized Mean Error Distance (NMED) for the layers provided.
+    
+    Args:
+        layer: Specify the layer number for 3x3 Convolutional layers of DNN
+
+    Returns:
+        Normalied Mean Error Distance (nmed)
+"""
+def calculate_nmed_layer(layer):
+    count = 0
+    EMAC = 0
+    NMED = 0
+    
+    for i in range(0, SVERILOG_BATCH_COUNT):
+        VERSION = "exact"
+        predictions_bin = []
+        with open(
+            ROOT_DIR
+            + "mult{}_{}in_layer{}_out.txt".format(VERSION, i, layer)
+        ) as pfile:
+            predictions_bin = pfile.readlines()
+
+        predictions_exact = twoscomp_to_decimal(predictions_bin, 8)
+
+        VERSION = "approx_lastlayer"
+        predictions_bin2 = []
+        with open(
+            ROOT_DIR
+            + "mult{}_{}in_layer{}_out.txt".format(VERSION, i, layer)
+        ) as pfile:
+            predictions_bin2 = pfile.readlines()
+
+        predictions_approx = twoscomp_to_decimal(predictions_bin2, 8)
+
+        for i in range(0, len(predictions_exact)):
+            EMAC += abs(predictions_approx[i] - predictions_exact[i])
+            count += 1
+
+    mean_abs_difference = (EMAC / count)
+    #print("MAD: ", mean_abs_difference)
+    NMED = mean_abs_difference / (MAX_VALUE)
+    return NMED
+
+# Calculate and print NMED of each layer.
+if VERSION != "approx_lastlayer":
+    nmed1 = calculate_nmed_layer(0)
+    nmed2 = calculate_nmed_layer(1)
+    nmed3 = calculate_nmed_layer(2)
+    print("NMED Layer 1: ", nmed1)
+    print("NMED Layer 2: ", nmed2)
+    print("NMED Layer 3: ", nmed3)
+elif VERSION == "approx_lastlayer":
+    nmed3 = calculate_nmed_layer(2)
+    print("NMED Last Layer: ", nmed3)
+else:
+    print("Error - Please input correct arg[1] to calculate NMED")