def demo_toy_training():
animals = cap.plugin.toy.animal.load_animals()
test_samples = [
cap.plugin.toy.extra_animal.load_animals(
samples_type=TYPE_TEST_SAMPLE)[8]
]
features_size = len(animals[0].features)
model = SOM2D(
features_size,
max_nbh_size=9,
nbh_step_size=0.3,
map_rows=17,
map_cols=17,
)
model.train(animals)
model.load_visualize_samples(animals, test_samples)
model.visualize_term()
fig = plt.figure()
ax = plt.subplot2grid((2, 3), (0, 0), colspan=2, rowspan=2)
model.visualize_plt(
ax,
29,
plt_style={
0: 'r^',
1: 'b*',
},
)
plt.tight_layout()
plt.show()
def test_nbhs2D(self):
""" to check if the function can locate 2D neighborhoods """
self.init_test(self.current_func_name)
model = SOM2D(5, map_rows=8, map_cols=9, random_seed=TEST_SEED)
nbh_indices = list(model.nbhs(42, 3))
self.assertEqual(23 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(24 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(25 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(26 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(27 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(28 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(29 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(60 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(61 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(17 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(18 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(45 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(53 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(len(nbh_indices), 28,
'Incorrect number of neighborhoods')
nbh_indices = list(model.nbhs(42, 4))
self.assertEqual(10 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(20 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(21 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(22 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(23 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(46 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(len(nbh_indices), 42,
'Incorrect number of neighborhoods')
nbh_indices = list(model.nbhs(42, 3.9))
self.assertEqual(10 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(20 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(21 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(22 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(23 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(46 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(len(nbh_indices), 40,
'Incorrect number of neighborhoods')
nbh_indices = list(model.nbhs(29, 1))
self.assertEqual(28 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(29 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(30 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(21 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(37 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(len(nbh_indices), 5,
'Incorrect number of neighborhoods')
nbh_indices = list(model.nbhs(32, 0))
self.assertEqual(32 in nbh_indices, True, 'Incorrect neighborhood')
self.assertEqual(33 in nbh_indices, False, 'Incorrect neighborhood')
self.assertEqual(len(nbh_indices), 1,
'Incorrect number of neighborhoods')
def test_init_default(self):
""" to check if SOM2D initialize correctly (default) """
self.init_test(self.current_func_name)
model = SOM2D(5)
self.assertEqual(model.features_size, 5,
'Incorrect number of features')
self.assertEqual(model.map_size, DFLT_MAP_ROWS * DFLT_MAP_COLS,
'Incorrect size of model mapping')
self.assertEqual(model.weight_step_size, DFLT_WEIGHT_STEP_SIZE,
'Incorrect step size')
self.assertEqual(model.max_nbh_size, DFLT_MAX_NBH_SIZE,
'Incorrect maximum number of neighborhood')
def test_from_grid(self):
""" to check if SOM2D can transform 2D positions to indices """
self.init_test(self.current_func_name)
model = SOM2D(
5,
map_rows=7,
map_cols=3,
)
test_row = 3
test_col = 2
test_idx = model.from_grid(test_row, test_col)
self.assertEqual(test_idx, 17, 'Incorrect idx')
test_row = 4
test_col = 1
test_idx = model.from_grid(test_row, test_col)
self.assertEqual(test_idx, 11, 'Incorrect idx')
def test_to_from_grid(self):
""" to check if SOM2D can transform between 2D positions to indices """
self.init_test(self.current_func_name)
model = SOM2D(
5,
map_rows=6,
map_cols=2,
)
test_row = 3
test_col = 1
test_idx = model.from_grid(test_row, test_col)
out_row, out_col = model.to_grid(test_idx)
self.assertEqual(out_row, 3, 'Incorrect 2D row')
self.assertEqual(out_col, 1, 'Incorrect 2D col')
test_idx = 5
test_row, test_col = model.to_grid(test_idx)
out_idx = model.from_grid(test_row, test_col)
self.assertEqual(out_idx, 5, 'Incorrect idx')
def test_to_grid(self):
""" to check if SOM2D can transform indices to 2D positions """
self.init_test(self.current_func_name)
model = SOM2D(
5,
map_rows=5,
map_cols=4,
)
test_row, test_col = model.to_grid(5)
self.assertEqual(test_row, 0, 'Incorrect 2D row')
self.assertEqual(test_col, 1, 'Incorrect 2D col')
test_row, test_col = model.to_grid(9)
self.assertEqual(test_row, 4, 'Incorrect 2D row')
self.assertEqual(test_col, 1, 'Incorrect 2D col')
test_row, test_col = model.to_grid(11)
self.assertEqual(test_row, 1, 'Incorrect 2D row')
self.assertEqual(test_col, 2, 'Incorrect 2D col')
test_row, test_col = model.to_grid(19)
self.assertEqual(test_row, 4, 'Incorrect 2D row')
self.assertEqual(test_col, 3, 'Incorrect 2D col')
def som2d(
training_samples,
test_samples=[],
visualize_params=[],
out_folder=None,
map_rows=DFLT_MAP_ROWS,
map_cols=DFLT_MAP_COLS,
weight_step_size=DFLT_WEIGHT_STEP_SIZE,
nbh_step_size=DFLT_NBH_STEP_SIZE,
max_nbh_size=DFLT_MAX_NBH_SIZE,
random_seed=DFLT_SEED,
):
features_size = len(training_samples[0].features)
model = SOM2D(
features_size,
map_rows=map_rows,
map_cols=map_cols,
weight_step_size=weight_step_size,
nbh_step_size=nbh_step_size,
max_nbh_size=max_nbh_size,
random_seed=random_seed,
)
#train and load sample for visualize
model.train(training_samples)
visualize_samples = []
for sample in training_samples:
visualize_samples.append(sample)
for sample in test_samples:
visualize_samples.append(sample)
model.load_visualize_samples(visualize_samples)
pdf_font = {'family': 'monospace', 'size': 3}
matplotlib.rc('font', **pdf_font)
#prepare text to visualize training attributes
training_samples_size = len(training_samples)
test_samples_size = len(test_samples)
iterations = int(ceil(float(model.max_nbh_size) / model.nbh_step_size))
col1_txt_fmt = "{caption:<28}:{value:>7}"
col1_txt = []
col1_txt.append(
col1_txt_fmt.format(caption="number of training samples",
value=training_samples_size))
col1_txt.append(
col1_txt_fmt.format(caption="number of test samples",
value=test_samples_size))
col1_txt.append(
col1_txt_fmt.format(caption="features size",
value=model.features_size))
col1_txt.append(
col1_txt_fmt.format(caption="training iterations", value=iterations))
col2_txt_fmt = "{caption:<24}:{value:>12}"
col2_txt = []
col2_txt.append(
col2_txt_fmt.format(caption="map rows", value=model.map_rows))
col2_txt.append(
col2_txt_fmt.format(caption="map cols", value=model.map_cols))
col2_txt.append(
col2_txt_fmt.format(caption="max neighborhod size",
value=model.max_nbh_size))
col2_txt.append(
col2_txt_fmt.format(caption="neighborhood step size",
value=model.nbh_step_size))
col2_txt.append(
col2_txt_fmt.format(caption="random seed", value=model.random_seed))
#generate individual reports
fig_rows = 1
fig_cols = 1
legend_width = 1
description_height = 1
fig_width = 2
fig_height = 2
plt_rows = fig_rows * fig_height + description_height
plt_cols = (fig_width + legend_width) * fig_cols
plt_txt_size = int(ceil(12 / fig_rows))
desc_txt_size = 10 - fig_rows
fig = plt.figure()
eps_file_names = []
idx = 0
for params in visualize_params:
fig_col = (idx % fig_cols) * (fig_width + legend_width)
fig_row = ((idx // fig_cols) * fig_height) + description_height
plt.subplot2grid((1, 1), (0, 0))
ax = plt.subplot2grid(
(plt_rows, plt_cols),
(fig_row, fig_col),
colspan=fig_width,
rowspan=fig_height,
)
if params['type'] == 'terminal':
out_file = os.path.join(out_folder, 'terminal_out.txt')
out_term = model.visualize_term(
txt_width=params['txt_width'],
out_file=out_file,
)
model.visualize_sample_name(ax, txt_size=plt_txt_size)
eps_file_name = os.path.join(out_folder, 'sample_names.eps')
elif params['type'] == 'scatter':
prop_name = params['prop_name']
model.visualize_plt(
ax,
prop_name,
params['plt_style'],
txt_size=plt_txt_size,
)
eps_file_name = os.path.join(out_folder,
'scatter_' + prop_name + '.eps')
elif params['type'] == 'debugging contour text':
prop_name = params['prop_name']
model.debugging_contour_txt(
ax,
prop_name,
txt_size=plt_txt_size,
)
eps_file_name = os.path.join(out_folder,
'dbg_contour_' + prop_name + '.eps')
elif params['type'] == 'debugging contour filter':
prop_name = params['prop_name']
model.debugging_contour_filter(
ax,
prop_name,
min_cutoff=params['min_cutoff'],
max_cutoff=params['max_cutoff'],
txt_size=plt_txt_size,
)
eps_file_name = os.path.join(
out_folder, 'dbg_contour_filter_' + prop_name + '.eps')
elif params['type'] == 'contour':
prop_name = params['prop_name']
out_plt = model.visualize_contour(
ax,
prop_name,
min_cutoff=params['min_cutoff'],
max_cutoff=params['max_cutoff'],
txt_size=plt_txt_size,
)
cbaxes = plt.subplot2grid(
(plt_rows, plt_cols * 2),
(fig_row, (fig_col + 2) * 2),
rowspan=fig_height,
)
plt.colorbar(out_plt, cax=cbaxes)
eps_file_name = os.path.join(out_folder,
'contour_' + prop_name + '.eps')
else:
continue
ax = plt.subplot2grid((plt_rows, plt_cols), (0, 0),
colspan=fig_cols * (fig_width + legend_width))
model.visualize_txt(ax, col1_txt, col2_txt, txt_size=desc_txt_size)
fig.savefig(eps_file_name, bbox_inches='tight', pad_inches=0.1)
eps_file_names.append(eps_file_name)
#generate summary pdf report
fig_rows = 2
fig_cols = 3
legend_width = 1
description_height = 1
fig_width = 2
fig_height = 2
plt_rows = fig_rows * fig_height + description_height
plt_cols = (fig_width + legend_width) * fig_cols
plt_txt_size = int(ceil(12 / fig_rows))
desc_txt_size = 12 - fig_rows
fig = plt.figure()
idx = 0
#plot figures
for params in visualize_params:
fig_col = (idx % fig_cols) * (fig_width + legend_width)
fig_row = ((idx // fig_cols) * fig_height) + description_height
ax = plt.subplot2grid(
(plt_rows, plt_cols),
(fig_row, fig_col),
colspan=fig_width,
rowspan=fig_height,
)
if params['type'] == 'terminal':
out_file = os.path.join(out_folder, 'terminal_out.txt')
out_term = model.visualize_term(
txt_width=params['txt_width'],
out_file=out_file,
)
model.visualize_sample_name(ax, txt_size=plt_txt_size)
elif params['type'] == 'scatter':
out_plt = model.visualize_plt(
ax,
params['prop_name'],
params['plt_style'],
txt_size=plt_txt_size,
)
elif params['type'] == 'debugging contour filter':
out_plt = model.debugging_contour_filter(
ax,
params['prop_name'],
min_cutoff=params['min_cutoff'],
max_cutoff=params['max_cutoff'],
txt_size=plt_txt_size,
)
elif params['type'] == 'contour':
out_plt = model.visualize_contour(
ax,
params['prop_name'],
min_cutoff=params['min_cutoff'],
max_cutoff=params['max_cutoff'],
txt_size=plt_txt_size,
)
cbaxes = plt.subplot2grid(
(plt_rows, plt_cols * 2),
(fig_row, (fig_col + 2) * 2),
rowspan=fig_height,
)
plt.colorbar(out_plt, cax=cbaxes)
elif params['type'] == 'debugging contour text':
out_plt = model.debugging_contour_txt(
ax,
params['prop_name'],
txt_size=plt_txt_size,
)
idx += 1
ax = plt.subplot2grid((plt_rows, plt_cols), (0, 0),
colspan=fig_cols * (fig_width + legend_width))
out_plt = model.visualize_txt(ax,
col1_txt,
col2_txt,
txt_size=desc_txt_size)
plt.tight_layout()
summary_pdf_file_name = os.path.join(out_folder, 'summary.pdf')
fig.savefig(summary_pdf_file_name, bbox_inches='tight', pad_inches=0.1)
return {
"summary file": summary_pdf_file_name,
"eps reports": eps_file_names,
"terminal file": out_term,
}