lettersim-ot-rsa/fig_code/99_illustrate_control_rdms.py

# %%
# imports

import os.path as op

import pandas as pd
import numpy as np
from string import ascii_lowercase

chars = [*ascii_lowercase, 'ä', 'ö', 'ü', 'ß']

import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import make_axes_locatable

fig_font_size = 8
fig_font_size_small = 8

plt.rcParams.update({
    "text.usetex": False,
    "font.family": "Helvetica",
    'font.size': fig_font_size
})

def rankify_vec(vec):
    # rank values in a vector
    # (ties have their ranks averaged, as in R)
    rank_vec = vec.argsort().argsort() + 1
    rank_vec = rank_vec.astype(float)
    for uv in np.unique(vec):
        uv_idx = vec==uv
        if np.sum(uv_idx)>1:
            rank_vec[uv_idx] = np.mean(rank_vec[uv_idx])
    return rank_vec

def rankify_sq_mat(mat):
    lwr = mat[np.tril_indices(n=mat.shape[0], k=-1)]
    lwr_rank = rankify_vec(lwr)
    mat_rank = np.zeros(mat.shape)
    mat_rank[np.tril_indices(n=mat.shape[0], k=-1)] = lwr_rank
    mat_rank[np.triu_indices(n=mat.shape[0], k=0)] = mat_rank.T[np.triu_indices(n=mat.shape[0], k=0)]
    mat_rank[np.diag_indices(n=mat_rank.shape[0])] = np.nan
    return mat_rank

#%matplotlib qt

# %%
# import the features

sz_feat_df = pd.read_csv( op.join('..', 'stim_sim', 'complexity', 'complexity_features.csv') )
f_feat_df = pd.read_csv( op.join('..', 'stim_sim', 'frequency', 'frequency_features.csv') )
ph_feat_df = pd.read_csv( op.join('..', 'stim_sim', 'phonology', 'dominant_phonemes_features.csv') )
phn_feat_df = pd.read_csv( op.join('..', 'stim_sim', 'phonology', 'letter_names_features.csv') )

# %%
# import the RDMs

sz_df = pd.read_csv( op.join('..', 'stim_sim', 'complexity', 'complexity.csv') )
sz_mat = np.load( op.join('..', 'stim_sim', 'complexity', 'complexity.npy') )

f_df = pd.read_csv( op.join('..', 'stim_sim', 'frequency', 'frequency.csv') )
f_mat = np.load( op.join('..', 'stim_sim', 'frequency', 'frequency.npy') )

ph_df = pd.read_csv( op.join('..', 'stim_sim', 'phonology', 'dominant_phonemes.csv') )
ph_mat = np.load( op.join('..', 'stim_sim', 'phonology', 'dominant_phonemes.npy') )

phn_df = pd.read_csv( op.join('..', 'stim_sim', 'phonology', 'letter_names.csv') )
phn_mat = np.load( op.join('..', 'stim_sim', 'phonology', 'letter_names.npy') )

# %%
# build the plot

fig, axs = plt.subplots(4, 4, figsize=(6, 4.8))
fig.subplots_adjust(left=None, bottom=None, right=None, top=None, wspace=0.75, hspace=0.55)

# titles
# axs[0, 0].set_ylabel('Letter Size\n(Pixel Sum)', rotation='horizontal', va='center')
# axs[1, 0].set_ylabel('Letter Frequency\n(SUBTLEX-DE Count)')
# axs[2, 0].set_ylabel('Dominant Phonemes\n(PanPhon Features)')
# axs[3, 0].set_ylabel('Letter Name Phonemes\n(PanPhon Features)')

titles = ['Letter Size\n(Pixel Sum)',
          'Letter\nFrequency\n(SUBTLEX-DE\nCount)',
          'Dominant\nPhonemes\n(PanPhon\nFeatures)',
          'Letter Name\nPhonemes\n(PanPhon\nFeatures)']

for i, t in enumerate(titles):
    axs[i, 0].text(0.75, 0.5, t,
          transform=axs[i, 0].transAxes,
          size=fig_font_size,
          verticalalignment='center',
          horizontalalignment='center')
    axs[i, 0].axis('off')

axs[0, 1].set_title('Features', fontsize=fig_font_size)
axs[0, 2].set_title('Raw RDMs', fontsize=fig_font_size)
axs[0, 3].set_title('Rank RDMs', fontsize=fig_font_size)

# size
axs[0, 1].bar(sz_feat_df.char, sz_feat_df.pixel_sum, width=1.0, color='black')
axs[0, 1].set_yticks([0, 2500, 5000])

sz_rdm_im = axs[0, 2].imshow(sz_mat, interpolation='none')

divider = make_axes_locatable(axs[0, 2])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(sz_rdm_im, cax=cax, orientation='vertical', ticks=[0, 2000, 4000])

sz_rank_im = axs[0, 3].imshow(rankify_sq_mat(sz_mat), interpolation='none')

divider = make_axes_locatable(axs[0, 3])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(sz_rank_im, cax=cax, orientation='vertical', ticks=[1, 435])

# frequency
axs[1, 1].bar(f_feat_df.char, f_feat_df.freq, width=1.0, color='black')
axs[1, 1].set_yticks(np.array([0, 0.5, 1, 1.5])*1e7)

f_rdm_im = axs[1, 2].imshow(f_mat, interpolation='none')

divider = make_axes_locatable(axs[1, 2])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(f_rdm_im, cax=cax, orientation='vertical', ticks=np.array([0, 0.4, 0.8, 1.2])*1e7)

f_rank_im = axs[1, 3].imshow(rankify_sq_mat(f_mat), interpolation='none')

divider = make_axes_locatable(axs[1, 3])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(f_rank_im, cax=cax, orientation='vertical', ticks=[1, 435])

# dominant phonemes
ph_feat_mat = ph_feat_df.loc[:, ph_feat_df.columns!='char'].to_numpy().T

ph_im = axs[2, 1].imshow(ph_feat_mat, interpolation='none', aspect='auto', cmap='coolwarm', vmin=-1, vmax=1)
axs[2, 1].set_yticks([0, 5, 11, 17, 23])
axs[2, 1].set_yticklabels([1, 6, 12, 18, 24])

divider = make_axes_locatable(axs[2, 1])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(ph_im, cax=cax, orientation='vertical')

ph_rdm_im = axs[2, 2].imshow(ph_mat, interpolation='none')

divider = make_axes_locatable(axs[2, 2])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(ph_rdm_im, cax=cax, orientation='vertical', ticks=[0, 0.4, 0.8, 1.2])

ph_rank_im = axs[2, 3].imshow(rankify_sq_mat(ph_mat), interpolation='none', vmin=1, vmax=435)

divider = make_axes_locatable(axs[2, 3])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(ph_rank_im, cax=cax, orientation='vertical', ticks=[1, 435])

# letter names
phn_feat_mat = phn_feat_df.loc[:, phn_feat_df.columns!='char'].to_numpy().T

phn_im = axs[3, 1].imshow(phn_feat_mat, interpolation='none', aspect='auto', cmap='coolwarm', vmin=-1, vmax=1)
axs[3, 1].set_yticks([0, 5, 11, 17, 23])
axs[3, 1].set_yticklabels([1, 6, 12, 18, 24])

divider = make_axes_locatable(axs[3, 1])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(phn_im, cax=cax, orientation='vertical')

phn_rdm_im = axs[3, 2].imshow(phn_mat, interpolation='none')

divider = make_axes_locatable(axs[3, 2])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(phn_rdm_im, cax=cax, orientation='vertical', ticks=[0, 0.5, 1])

phn_rank_im = axs[3, 3].imshow(rankify_sq_mat(phn_mat), interpolation='none')

divider = make_axes_locatable(axs[3, 3])
cax = divider.append_axes('right', size='5%', pad=0.05)
fig.colorbar(phn_rank_im, cax=cax, orientation='vertical', ticks=[1, 435])

# axes labels for characters
for i in range(4):
    axs[i, 1].set_xticks([0, 14, 29])
    axs[i, 1].set_xticklabels(['a', '...', 'ß'])
    xticks = axs[i, 1].xaxis.get_major_ticks()
    xticks[1].get_children()[0].set_visible(False)

for i in range(4):
    for j in range(2, 4):
        axs[i, j].set_xticks([0, 14, 29])
        axs[i, j].set_yticks([0, 14, 29])
        axs[i, j].set_xticklabels(['a', '...', 'ß'])
        axs[i, j].set_yticklabels(['a', '...', 'ß'])
        xticks = axs[i, j].xaxis.get_major_ticks()
        xticks[1].get_children()[0].set_visible(False)
        yticks = axs[i, j].yaxis.get_major_ticks()
        yticks[1].get_children()[0].set_visible(False)
        yticks[1].get_children()[3].set_rotation(90)


fig.savefig(op.join('..', 'fig', 'illustrate_controls.pdf'))
fig.savefig(op.join('..', 'fig', 'illustrate_controls.png'))