def plt_anyfeature_mean(ftr_name):
import matplotlib.colors as mcolors
color = np.zeros((m, n))
size = np.ones_like(hits)
for x in range(m):
for y in range(n):
color[x, y] = data[ftr_name].iloc[wmix[x, y]].mean()
vmax = data[ftr_name].max()
vmin = data[ftr_name].min()
color = np.nan_to_num(color)
cbmin = color.min()
cbmax = color.max()
color = (color - color.min()) / (color.max() - color.min())
fig, ax = plt.subplots(1, 1, figsize=(m / 2, n / 2))
cmap = 'inferno_r'
map_plot(ax,
dist,
color,
m,
n,
size=size,
scale=4,
title=ftr_name + ' mean',
cmap=cmap)
norm = mcolors.Normalize(vmin=vmax, vmax=vmin)
sm = plt.cm.ScalarMappable(cmap=cmap, norm=norm)
sm.set_array([])
fig.subplots_adjust(right=0.75, left=0.1)
cbar1 = fig.add_axes([0.8, 0.25, 0.05, 0.45])
cb = plt.colorbar(sm, cax=cbar1)
cb.ax.tick_params(labelsize='x-small')
def plt_features(ftr_name):
ftr = feat[case].index(ftr_name)
size = np.ones_like(hits)
WW = W.reshape(m * n, bneck)
if pca and not acode:
WW = pcomp.inverse_transform(WW)
if acode:
WW = ae.decode(torch.Tensor(WW)).detach().numpy()
WW = scaler.inverse_transform(WW)
WW = WW.reshape(m, n, nfeat)
color = WW[:, :, ftr]
cmin = color.min()
cmax = color.max()
color = (color - cmin) / (cmax - cmin)
fig, ax = plt.subplots(1, 1)
map_plot(ax, dist, color, m, n, size=size, scale=8, cmap='inferno_r')
plt.title(ftr_name)
def plt_mapdatamean(K):
color = np.zeros((m, n))
size = hits
for x in range(m):
for y in range(n):
color[x, y] = data[K].iloc[wmix[x, y]].mean()
color = np.nan_to_num(color)
cbmin = color.min()
cbmax = color.max()
color = (color - cbmin) / (cbmax - cbmin)
fig, ax = plt.subplots(1, 1)
cmap = plt.cm.get_cmap('jet_r', 5)
map_plot(ax,
dist,
color,
m,
n,
size=size,
scale=6,
cmap=cmap,
lcolor='black')
data_added = []
for i in range(len(data)):
x, y = som.winner(data[i])
data_added.append(mapdata[x, y])
data_added = np.array(data_added)
return data_added
def som_addinfo(som, df, data, mapdata, mapname):
assert (som.get_weights().shape[0] == mapdata.shape[0])
assert (som.get_weights().shape[1] == mapdata.shape[1])
data_added = som_adddata(som, data, mapdata)
df[mapname] = data_added
return df
if __name__ == "__main__":
from matplotlib_hex_map import matplotlib_hex_map as map_plot
m = 7
n = 9
model, data = som_colortest(m, n)
d = som_distances(model)
hits = som_hits(model, data, m, n, log=False)
wmi = model.win_map_index(data)
color = model.get_weights()[:, :, :3]
color = (color - color.min()) / (color.max() - color.min())
map_plot(d, color, m, n, size=hits, scale=3)
# plt.close('all')
color = W.sum(axis=2)
cmin = color.min() #np.min(x, axis=0)
cmax = color.max() #np.max(x, axis=0)
color = (color - cmin) / (cmax - cmin)
if plot_hitmap:
size = hits # np.ones_like(hits)
fig, ax = plt.subplots(1, 1)
map_plot(ax,
dist,
color,
m,
n,
size=size,
scale=3,
cmap='inferno_r',
lcolor='black')
plt.title('Hit map')
if plot_neighbors:
f = lambda p, q: p - 0.5 if (q % 2 == 0) else p
i = f(px, py)
j = py
plt.plot([i, i + 0.5], [j * 0.75, j * 0.75 + 0.75], 'k-')
plt.plot([i, i + 1], [j * 0.75, j * 0.75], 'k-')
plt.plot([i, i + 0.5], [j * 0.75, j * 0.75 - 0.75], 'k-')
plt.plot([i, i - 0.5], [j * 0.75, j * 0.75 - 0.75], 'k-')