Merge pull request #380 from ldbc/feature/estimation-script

tuning/sf-prediction.py

@@ -0,0 +1,75 @@
+"""
+FILE: sf-prediction.py
+DATE: 24 February 2022
+
+DESC: Predict the number of nodes for different scale factors.
+      The known values for each scale factor are:
+     _____________________________________________________________________
+    | Scale Factor    |    1  |   3   |   10  |   30   |  100   |  300    |
+    ----------------------------------------------------------------------
+    | Nr. of Persons  | 10620 | 25870 | 70800 | 175950 | 487700 | 1230500 |
+    ----------------------------------------------------------------------
+
+     ___________________________________________________________
+    | Scale Factor    |   1000  |  3000   |  10000   |  30000   |
+    -------------------------------------------------------------
+    | Nr. of Persons  | 3505000 | 9232000 | 27200000 | 77000000 |
+    ------------------------------------------------------------  
+
+    The values are approximated using a polynomial with degree 3 for scale
+    factors untill 3000 with the nr. of persons and scale factors scaled 
+    with natural logarithm. For scale factors larger than 3000, a 5th degree
+    polynomial is used without log-scale.
+"""
+
+import numpy as np
+
+def approximate_small_sf(sf, num_persons, sf_new):
+    """
+    :param: sf (List):          List of scale factors
+    :param: num_persons (List): List of number of persons aligned with the
+                                scale factor list
+    :param: sf_new (List):      The scale factors to predict
+
+    :return: List of predicted number of persons
+    """
+    coeffs = np.polyfit(np.log(sf), np.log(num_persons), deg=3)
+    poly = np.poly1d(coeffs)
+    yfit = lambda x: np.exp(poly(np.log(x)))
+    return np.array(yfit(sf_new))
+
+def approximate_large_sf(sf, num_persons, sf_new):
+    """
+    :param: sf (List):          List of scale factors
+    :param: num_persons (List): List of number of persons aligned with the
+                                scale factor list
+    :param: sf_new (List):      The scale factors to predict
+
+    :return: List of predicted number of persons
+    """
+    coeffs = np.polyfit(sf, num_persons, deg=5)
+    poly = np.poly1d(coeffs)
+    return np.array(poly(sf_new))
+
+if __name__ == "__main__":
+    sf = [1, 3, 10, 30, 100, 300, 1000, 3000, 10000, 30000]
+    num_persons = [
+        10620, 25870, 70800, 175950, 487700, 1230500, 3505000, 9232000,
+        27200000, 77000000
+    ]
+    sf_new = [1, 3, 10, 30, 100, 300, 1000, 3000, 10000, 30000, 100000]
+
+    predicted_large_sf = approximate_large_sf(sf, num_persons, sf_new)
+    predicted_small_sf = approximate_small_sf(sf, num_persons, sf_new)
+
+    print("Number of Persons")
+    print(num_persons)
+    print("Polyfit Large")
+    print(predicted_large_sf)
+    print("Polyfit Small")
+    print(predicted_small_sf)
+
+    print("Error Polyfit Large")
+    print(np.abs(predicted_large_sf[:10] - num_persons))
+    print("Error Polyfit Small")
+    print(np.abs(predicted_small_sf[:10] - num_persons))