Export 1% relerr

utils/python/linelast_cwtrain.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: MIT
 
 import numpy as np
+from scipy.interpolate import interp1d
 import h5py
 import os
 import matplotlib.pyplot as plt
@@ -153,6 +154,7 @@ def get_results(samples, prefix):
 def solve_time_scaling_plot(samples, res, scale_prefix, plt_name = "scaling.png"):
     plt.plot(samples, res[:,0], label='FOM')
     plt.plot(samples, res[:,1], label='ROM')
+    plt.xscale('log')
     plt.xlabel(scale_prefix)
     plt.yscale('log')
     plt.ylabel("Solve time [s]")
@@ -164,6 +166,7 @@ def solve_time_scaling_plot(samples, res, scale_prefix, plt_name = "scaling.png"
 
 def relerr_scaling_plot(samples, res, scale_prefix, plt_name = "scaling.png"):
     plt.plot(samples, res[:,2], label='Relative error')
+    plt.xscale('log')
     plt.xlabel(scale_prefix)
     plt.yscale('log')
     plt.ylabel("Relative error [-]")
@@ -174,18 +177,34 @@ def relerr_scaling_plot(samples, res, scale_prefix, plt_name = "scaling.png"):
 
 def speedup_scaling_plot(samples, res, scale_prefix, plt_name = "scaling.png"):
     plt.plot(samples, res[:,3])
+    plt.xscale('log')
     plt.xlabel(scale_prefix)
     plt.ylabel("Speedup factor [-]")
 
     plt.tight_layout()
     plt.savefig("speedup_" + plt_name, dpi=300)
     plt.clf()
 
+def export_opt_val(filename, opt_nb_a, opt_nb_i, opt_speedup):
+    f = open(filename, "w")
+    out_txt = "Optimal number of bases (interpolated) is: " + str(round(float(opt_nb_a), 4)) + " bases\nOptimal number of bases (rounded) is: " + str(opt_nb_i) + " bases\nSpeedup at rounded number of bases is: " + str(round(float(opt_speedup), 4)) + " x"
+    f.write(out_txt)
+    f.close()
+
 def create_scaling_plot(samples, res, scale_prefix, plt_name = "plot.png"):
     plt.rc('axes', labelsize=14)
+    ferr = interp1d(samples, res[:,2]) # Rel err
+    ferr_i = interp1d(res[:,2], samples) # Inverse correlation
+    x_star_a = ferr_i(1e-2) # Analytical x_star
+    x_star_i = np.ceil(x_star_a) # Next integer
+
+    fspeed = interp1d(samples, res[:,3]) # speedup factor
+    opt_speedup = fspeed(x_star_i)
+
     solve_time_scaling_plot(samples, res, scale_prefix, plt_name)
     relerr_scaling_plot(samples, res, scale_prefix, plt_name)
     speedup_scaling_plot(samples, res, scale_prefix, plt_name)
+    export_opt_val("opt_vals.txt", x_star_a, x_star_i, opt_speedup)
 
 def get_nr(txt, split_txt = 'comparison'):
     return int(txt.split('.')[0].split(split_txt)[1])