05_plot_test.py

"""
Plot the performance on the test set as in Figures 6, 12, 18.
"""
# Author: Arturs Berzins <berzins@cats.rwth-aachen.de>
# License: BSD 3 clause

import config
import utils
import matplotlib
from matplotlib import pyplot
import numpy as np
from matplotlib.ticker import FixedLocator, NullFormatter

preamble = (
    r'\usepackage{amsmath}'
    r'\usepackage{amssymb}'
    r'\newcommand{\vekt}[1]{\mbox{$\boldsymbol{#1}$}}'
    )
            
matplotlib.rcParams.update({
    "pgf.texsystem": "pdflatex",
    'font.family': 'serif',
    'text.usetex': True,
    'text.latex.preamble': preamble,
    'pgf.preamble': preamble,
    'pgf.rcfonts': False,
    'font.size': 8
})
#https://timodenk.com/blog/exporting-matplotlib-plots-to-latex/

## In case we don't want to plot all L for visibility
L_max = [20,20,20]
dataset = 'test'
df = utils.load_error_table(dataset)
fig, axes = pyplot.subplots(1,len(config.components))
fig.suptitle('')
legends = []
cmap = pyplot.get_cmap('tab10')
keys = [F'eps_pod{m.lower()}_sq' for m in list(utils.models.keys()) + ['']]

for idx_ax, component in enumerate(config.components):
    L = config.num_basis[component]
    ls = [l for l in range(L+1)]
    
    df_filtered = df.loc[ (df['component']==component)]
    
    markers = ['1','2','3','4']
    
    for i, key in enumerate(keys):
        mean_sq = np.array([ df_filtered.loc[(df_filtered['l']==l)][key].mean() for l in ls ])
        mean    = np.array([(df_filtered.loc[(df_filtered['l']==l)][key]**0.5).mean() for l in ls ])
        
        rmse = np.sqrt(mean_sq)
        axes[idx_ax].plot(ls, rmse, color=cmap(i), marker=markers[i], markersize=6, linewidth=1, markevery=5)
    
    axes[idx_ax].set_yscale('log')
    axes[idx_ax].set_ylim([None,1])
    axes[idx_ax].set_xlim([0, L_max[idx_ax]])
    #axes[idx_ax].set_xlabel(r'$L_{}$'.format(component))
    axes[idx_ax].set_xlabel(r'$L$')
    axes[idx_ax].grid(which='major', axis='y', linestyle=':')
    axes[idx_ax].grid(which='major', axis='x', linestyle=':')
    axes[idx_ax].title.set_text(r'${}$'.format([r'\vekt{u}',r'p',r'T'][idx_ax]))
    
    # all 10 integer powers between min and max
    displayed_data = rmse[:L_max[idx_ax]]
    exp_min = np.floor(np.log10(np.min(displayed_data)))
    exp_max = 0#np.log10(np.max(rmse))
    axes[idx_ax].set_ylim([10**exp_min, 10**exp_max])
    # MAJOR
    exps_major = np.arange( exp_min, exp_max+1 )
    axes[idx_ax].yaxis.set_major_locator(FixedLocator(10**exps_major))
    
    # MINOR
    axes[idx_ax].yaxis.set_minor_formatter(NullFormatter())
    
    axes[idx_ax].yaxis.set_ticks_position('both')
    axes[idx_ax].tick_params(axis='y', direction="in", which='both')

labels = [r'${\varepsilon}' + r'_\text{{{}}}'.format(key) + r'(\mathbb{P}_{te})$' for key in [r'POD-RBF', 'POD-GPR', 'POD-ANN', 'POD'] ]
lines = axes[0].get_lines()
fig.legend(lines, labels, ncol = 4, loc="lower center")
fig.subplots_adjust(bottom=0.35, top=0.90, left=0.10, right=0.95, wspace=0.35, hspace=0.20)

fig.set_size_inches(w=6.3, h=2.3)
     
pyplot.show()