file=outfile)
print(f" \caption{{{caption}}}", file=outfile)
print(f" \label{{fig:{label}}}", file=outfile)
print(f"\end{{figure*}}", file=outfile)
if len(sys.argv) < 5:
print(f"USAGE: {sys.argv[0]} IMG SOM MAP TRANSFORM")
imgs = pu.ImageReader(sys.argv[1])
som = pu.SOM(sys.argv[2])
mapping = pu.Mapping(sys.argv[3])
trans = pu.Transform(sys.argv[4])
somset = pu.SOMSet(som, mapping, trans)
path = pu.PathHelper("Inspection")
outfile = open("latex_imgs.tex", "w")
### SOM 2D histogram ###
som_counts(somset, True)
plt.savefig(f"som_stats.png")
plt.clf()
caption = f"The total number of images within the {imgs.data.shape[0]} input images mapped to each neuron in the SOM."
latex_figure("som_stats.png", caption, "som_stats", outfile)
print("", file=outfile)
########################
### Euclidean Distance ###
dist_hist(somset, log=False)
plt.savefig(f"EucDist_hist_lin.png")
dist_hist(somset, log=True)
def neuron_img_comp(somset, imgs, sampler, outpath=""):
path = pu.PathHelper(outpath)
logfile = open(f"{path.path}/info.txt", "w")
bmu_ed = somset.mapping.bmu_ed()
img_shape = imgs.data.shape[-1]
neuron_shape = somset.som.neuron_shape[-1]
b1 = (neuron_shape - img_shape) // 2
b2 = b1 + img_shape
for neuron, ind in enumerate(sampler.points):
# Plot neuron
somset.som.plot_neuron(ind)
f1 = plt.gcf()
plt.xticks([])
plt.yticks([])
radio_img = somset.som[ind][0]
levels = np.linspace(0.25 * radio_img.max(), radio_img.max(), 4)
f1.axes[1].contour(radio_img, levels=levels, colors="white", linewidths=0.5)
f1.axes[0].set_xlim([b1, b2])
f1.axes[0].set_ylim([b2, b1])
for ax in f1.axes:
ax.axvline(neuron_shape / 2, c="r", ls="--", lw=1)
ax.axhline(neuron_shape / 2, c="r", ls="--", lw=1)
f1.savefig(f"{path.path}/neuron_{neuron}.png")
plt.close(f1)
# Plot images
matches = somset.mapping.images_with_bmu(ind)
dist = bmu_ed[matches]
idx1 = matches[np.argmin(dist)]
idx2 = matches[np.argsort(dist)[len(matches) // 2]]
for i, idx in enumerate([idx1, idx2]):
plot_image(
imgs,
idx=idx,
somset=somset,
apply_transform=True,
show_index=False,
grid=True,
)
f2 = plt.gcf()
plt.xticks([])
plt.yticks([])
for ax in f2.axes:
ax.axvline(img_shape / 2, c="r", ls="--", lw=1)
ax.axhline(img_shape / 2, c="r", ls="--", lw=1)
bmu_idx = somset.mapping.bmu(idx)
tkey = somset.transform.data[(idx, *bmu_idx)]
radio_img = imgs.data[idx, 0]
radio_img = pu.pink_spatial_transform(radio_img, tkey)
levels = np.linspace(0.25 * radio_img.max(), radio_img.max(), 4)
f2.axes[1].contour(radio_img, levels=levels, colors="white", linewidths=0.5)
f2.savefig(f"{path.path}/neuron_{neuron}_img{i}.png")
plt.close(f2)
# Print to a log file
print(f"Neuron {neuron}: {ind}", file=logfile)
print(f"Number in neuron: {len(matches)}", file=logfile)
print(f"img0 ind: {idx1}", file=logfile)
print(f"img1 ind: {idx2}", file=logfile)
print("------------------\n", file=logfile)
sns.scatterplot(data=neuron_cat[good_neurons],
x="Area_ratio",
y="Radio_area",
hue="good")
candidates = neuron_cat.Neuron_ID[mask].values
choices = np.random.choice(candidates, 45, replace=False)
print(choices)
plt.imshow(img_positions(choices, som.som_shape[0]))
# from sklearn.cluster import KMeans
# km = KMeans(30).fit(list(candidates))
# choices = [tuple(ki) for ki in km.cluster_centers_.astype(int)]
path = pu.PathHelper("Zoo_revised")
for neuron in choices:
plt.close("all")
som.plot_neuron(neuron)
plt.savefig(f"{path.path}/neuron_{neuron[0]}-{neuron[1]}.png")
doubles = [
(0, 8),
# (0, 28),
(0, 29),
# (1, 8),
(4, 27),
(6, 18),
(28, 9),
(30, 25),
(31, 4),
)
args = parser.parse_args()
return args
if __name__ == "__main__":
args = parse_args()
# Initialize the sample DataFrame, ImageReader, and SOM
# files = set_filenames(args.som_file, args.map_file, args.trans_file, args.outbase)
# df = pd.read_csv(args.sample_file)
imgs = pu.ImageReader(args.imbin_file)
somset = init_somset(args.som_file, args.map_file, args.trans_file)
path = pu.PathHelper(args.outpath)
# Only keep the records that preprocessed successfully
# df = df.loc[imgs.records].reset_index(drop=True)
# df = update_component_catalogue(df, somset)
som_counts(somset, show_counts=True)
plt.savefig(f"{path}/bmu_frequency.png")
# plot_image(imgs, df=df, somset=somset, show_bmu=True, apply_transform=True)
# dist_hist(df, labels=True)
# plt.savefig(f"{path}/euc_distance_hist.png")
# dist_hist_2d(df, somset, bins=100, loglog=False)
# plt.savefig(f"{path}/euc_distance_hist_2d.png", dpi=somset.som.som_shape[0] * 25)