def plot_discoveries_treemap(experiment_definition, repetition_id=0, experiment_statistics=None, data_filters=None, config=None, **kwargs):
default_config = dict(
random_seed = 0,
squarify = dict(
x = 0.0,
y = 0.0,
width = 100.0,
height = 100.0,
),
# global style config
global_layout = dict(
height=700,
width=700,
margin = dict(
l = 0,
r = 0,
b = 0,
t = 0
),
xaxis=dict(
autorange=True,
showgrid=False,
zeroline=False,
showline=False,
ticks='',
showticklabels=False
),
yaxis=dict(
autorange=True,
showgrid=False,
zeroline=False,
showline=False,
ticks='',
showticklabels=False
),
title = dict(
text = '',
font = dict(
color = "black",
size = 22,
family='Times New Roman'
)
),
hovermode='closest',
showlegend = True,
),
# Shapes style config
shapes = dict(
line = dict(
width = 2
),
layer = "below"
),
shapes_background_colors = ['rgb(0,0,0)', 'rgb(204,121,167)', 'rgb(0,114,178)'],
shapes_lines_colors = [ 'rgb(0,0,0)','rgb(154,71,117)', 'rgb(0,64,128)'],
# Images style config
margin_x = 1,
margin_y = 1,
images = dict (
xref= "x",
yref= "y",
sizex= 10,
sizey= 10,
opacity = 1,
xanchor= "center",
yanchor= "middle",
layer = "above"
),
images_transform = True,
images_transform_colormaps = [
create_colormap(np.array([[255,255,255], [119,255,255],[23,223,252],[0,190,250],[0,158,249],[0,142,249],[81,125,248],[150,109,248],[192,77,247],[232,47,247],[255,9,247],[200,0,84]])/255*8), #WBPR
create_colormap(np.array([[0,0,4],[0,0,8],[0,4,8],[0,8,8],[4,8,4],[8,8,0],[8,4,0],[8,0,0],[4,0,0]])), #BCYR
create_colormap(np.array([[0,2,0],[0,4,0],[4,6,0],[8,8,0],[8,4,4],[8,0,8],[4,0,8],[0,0,8],[0,0,4]])), #GYPB
create_colormap(np.array([[4,0,2],[8,0,4],[8,0,6],[8,0,8],[4,4,4],[0,8,0],[0,6,0],[0,4,0],[0,2,0]])), #PPGG
create_colormap(np.array([[4,4,6],[2,2,4],[2,4,2],[4,6,4],[6,6,4],[4,2,2]])), #BGYR
create_colormap(np.array([[4,6,4],[2,4,2],[4,4,2],[6,6,4],[6,4,6],[2,2,4]])), #GYPB
create_colormap(np.array([[6,6,4],[4,4,2],[4,2,4],[6,4,6],[4,6,6],[2,4,2]])), #YPCG
create_colormap(np.array([[8,8,8],[7,7,7],[5,5,5],[3,3,3],[0,0,0]]), is_marker_w=False), #W/B
create_colormap(np.array([[0,0,0],[3,3,3],[5,5,5],[7,7,7],[8,8,8]]))], #B/W
# Annotations style config
annotations = dict(
font = dict(
color = "#140054",
size = 18,
family='Times New Roman'
),
showarrow = False,
opacity=1.0,
bgcolor='rgb(255,255,255)'
)
)
config = ad.config.set_default_config(kwargs, config, default_config)
random.seed(config.random_seed)
# experiment_ids=[exp_def['id'] for exp_def in experiment_definitions]
# for experiment_id in experiment_ids:
# experiment_idx = experiment_ids.index(experiment_id)
# for repetition_idx in repetition_ids:
# experiment_statistics[experiment_id][repetition_idx] = ad.gui.jupyter.misc.load_statistics(os.path.join(experiment_definitions[experiment_idx]['directory'], 'repetition_{:06d}'.format(repetition_idx)))
experiment_id = experiment_definition['id']
#exp_idx = experiment_ids.index(experiment_id)
if repetition_id is None:
path = experiment_definition['directory']
else:
path = os.path.join(
experiment_definition['directory'],
'repetition_{:06d}'.format(int(repetition_id))
)
path = os.path.join(path, 'statistics')
final_observation_path = os.path.join(path, 'final_observation')
# Recover images (from png or zip)
all_images = []
if os.path.isdir(final_observation_path):
dir_content = os.listdir(final_observation_path)
for image_file_name in dir_content:
file_path = os.path.join(path, image_file_name)
if os.path.isfile(file_path) and '.png' in image_file_name:
item_id = image_file_name.split('.')[0]
item_id = int(item_id)
file = open(file_path, "rb")
image_PIL = Image.open(file)
if config.images_transform:
image_PIL = transform_image_from_colormap(image_PIL, config.images_transform_colormaps[0])
all_images.append(image_PIL)
elif zipfile.is_zipfile(final_observation_path + '.zip'):
zf = zipfile.ZipFile(final_observation_path + '.zip', 'r')
dir_content = zf.namelist()
for image_file_name in dir_content:
item_id = image_file_name.split('.')[0]
item_id = int(item_id)
with zf.open(image_file_name) as file:
image_PIL = Image.open(file)
if config.images_transform:
image_PIL = transform_image_from_colormap(image_PIL, config.images_transform_colormaps[0])
all_images.append(image_PIL)
else:
warnings.warn('No Images found under {!r}'.format(final_observation_path))
# Recover labels
filters_ids = list(data_filters.keys())
database_filtered_ids = dict()
for curr_filter_k, curr_filter_val in data_filters.items():
# filter
if curr_filter_val is None or not curr_filter_val:
img_ids = range(len(all_images))
else:
run_ids = ad.gui.jupyter.misc.filter_single_experiment_data(experiment_statistics[experiment_id], curr_filter_val, int(repetition_id))
img_ids = np.where(run_ids)[0]
database_filtered_ids[curr_filter_k] = list(img_ids)
values = []
for filter_id in filters_ids:
values.append(len(database_filtered_ids[filter_id])/len(all_images)*100)
sorted_indexes = np.argsort(-np.array(values))
values = np.array(values)[sorted_indexes]
filters_ids = np.array(filters_ids)[sorted_indexes]
config.shapes_background_colors = np.array(config.shapes_background_colors)[sorted_indexes]
config.shapes_lines_colors = np.array(config.shapes_lines_colors)[sorted_indexes]
normed = ad.gui.jupyter.squarify.normalize_sizes(values, config.squarify.width, config.squarify.height)
rects = ad.gui.jupyter.squarify.squarify(normed, config.squarify.x, config.squarify.y, config.squarify.width, config.squarify.height)
annotations = []
shapes = []
images = []
counter = 0
for r in rects:
# annotations layout
annotation_config = ad.config.set_default_config(
dict(
x = r['x']+(r['dx']/2),
y = r['y']+(r['dy']/2),
text = "<b>{}:<br>{:.1f}%<b>".format(filters_ids[counter],values[counter]),
),
config.annotations
)
annotations.append(annotation_config)
# shapes layout
shape_config = ad.config.set_default_config(
dict(
type = 'rect',
x0 = r['x'],
y0 = r['y'],
x1 = r['x']+r['dx'],
y1 = r['y']+r['dy'],
fillcolor = config.shapes_background_colors[counter],
line = dict(color = config.shapes_lines_colors[counter])
),
config.shapes
)
shapes.append(shape_config)
# images layout
x0 = r['x']
y0 = r['y']
w = r['dx']
h = r['dy']
n_cols = int((w - 2 * config.margin_x) // config.images.sizex)
space_x = (w - 2 * config.margin_x) / n_cols
centers_x = []
for j in range(n_cols):
centers_x.append(j * space_x + space_x / 2.0)
n_rows = int((h - 2 * config.margin_y) // config.images.sizey)
space_y = (h - 2 * config.margin_y) / n_rows
centers_y = []
for i in range(n_rows):
centers_y.append(i * space_y + space_y / 2.0)
list_of_random_items = random.sample(database_filtered_ids[filters_ids[counter]], n_rows*n_cols)
for i in range(n_rows):
for j in range(n_cols):
image_config = ad.config.set_default_config(
dict(
source = all_images[list_of_random_items[np.ravel_multi_index((i, j), dims=(n_rows,n_cols), order='F')]],
x = x0 + config.margin_x + centers_x[j],
y = y0 + config.margin_y + centers_y[i],
),
config.images)
images.append(image_config)
counter = counter + 1
if counter >= len(config.shapes_background_colors):
counter = 0
# append to global layout
global_layout = ad.config.set_default_config(
dict(
annotations = annotations,
shapes=shapes,
images=images
),
config.global_layout
)
figure = dict(data=[go.Scatter()],layout=global_layout)
plotly.offline.iplot(figure)
return figure
def get_default_gui_config():
default_config = ad.Config()
# general box layout config
default_config.box_layout = ad.Config()
default_config.box_layout.flex_flow = 'column'
default_config.box_layout.display = 'flex'
default_config.box_layout.align_items = 'stretch'
# Row 0: 2D visualisation type selector
default_config.visualisation_type_selector = ad.Config()
default_config.visualisation_type_selector.description = 'visualisation type'
default_config.visualisation_type_selector.options = [
'T-SNE', 'PCA', 'UMAP', '0-1', '0-2', '0-3', '0-4', '0-5', '0-6',
'0-7'
]
default_config.visualisation_type_selector.value = 'UMAP'
default_config.visualisation_type_selector.disabled = False
# Row 1: Canvas graph
default_config.canvas = ad.Config()
## default print properties for layout
multiplier = 2
pixel_cm_ration = 35
width_full = int(13.95 * pixel_cm_ration) * multiplier
width_half = int(13.95 / 2 * pixel_cm_ration) * multiplier
height_default = int(3.5 * pixel_cm_ration) * multiplier
top_margin = 0 * multiplier
left_margin = 0 * multiplier
right_margin = 0 * multiplier
bottom_margin = 0 * multiplier
font_family = 'Times New Roman'
font_size = 10 * multiplier
default_config.canvas.layout = dict(
font=dict(
family=font_family,
size=font_size,
),
updatemenus=[],
width=width_half, # in cm
height=height_default, # in cm
margin=dict(
l=left_margin, #left margin in pixel
r=right_margin, #right margin in pixel
b=bottom_margin, #bottom margin in pixel
t=top_margin, #top margin in pixel
),
legend=dict(
font=dict(
family=font_family,
size=font_size,
),
xanchor='right',
yanchor='bottom',
y=0.,
x=1.0,
),
showlegend=False)
## scatter default config
default_config.canvas_data = ad.Config()
default_config.canvas_data.mode = 'markers'
default_config.canvas_data.marker = dict(size=4.0, opacity=1.0)
## colors and symbols for markers
default_config.marker_colorlist = dict()
default_config.marker_colorlist['dead'] = 'rgb(0,0,0)'
default_config.marker_colorlist['animal'] = 'rgb(230,159,0)'
default_config.marker_colorlist['non-animal'] = 'rgb(0,158,115)'
default_config.marker_symbollist = dict()
default_config.marker_symbollist['dead'] = 'circle'
default_config.marker_symbollist['animal'] = 'diamond'
default_config.marker_symbollist['non-animal'] = 'square'
# Row 2: Image list widget for selected points
default_config.image_list_widget = ad.gui.jupyter.ImageListWidget.get_default_gui_config(
)
default_config.image_list_widget.image_items.layout.border = '1px solid gray' # add gray countour around images displayed
## colormap for images (BW --> colormap)
default_config.transform_images = True
default_config.transform_images_colormaps = [
create_colormap(
np.array([[255, 255, 255], [119, 255, 255], [23, 223, 252],
[0, 190, 250], [0, 158, 249], [0, 142, 249],
[81, 125, 248], [150, 109, 248], [192, 77, 247],
[232, 47, 247], [255, 9, 247], [200, 0, 84]]) / 255 *
8), #WBPR
create_colormap(
np.array([[0, 0, 4], [0, 0, 8], [0, 4, 8], [0, 8,
8], [4, 8, 4],
[8, 8, 0], [8, 4, 0], [8, 0, 0], [4, 0, 0]])), #BCYR
create_colormap(
np.array([[0, 2, 0], [0, 4, 0], [4, 6, 0], [8, 8,
0], [8, 4, 4],
[8, 0, 8], [4, 0, 8], [0, 0, 8], [0, 0, 4]])), #GYPB
create_colormap(
np.array([[4, 0, 2], [8, 0, 4], [8, 0, 6], [8, 0,
8], [4, 4, 4],
[0, 8, 0], [0, 6, 0], [0, 4, 0], [0, 2, 0]])), #PPGG
create_colormap(
np.array([[4, 4, 6], [2, 2, 4], [2, 4, 2], [4, 6, 4],
[6, 6, 4], [4, 2, 2]])), #BGYR
create_colormap(
np.array([[4, 6, 4], [2, 4, 2], [4, 4, 2], [6, 6, 4],
[6, 4, 6], [2, 2, 4]])), #GYPB
create_colormap(
np.array([[6, 6, 4], [4, 4, 2], [4, 2, 4], [6, 4, 6],
[4, 6, 6], [2, 4, 2]])), #YPCG
create_colormap(np.array([[8, 8, 8], [7, 7, 7], [5, 5, 5],
[3, 3, 3], [0, 0, 0]]),
is_marker_w=False), #W/B
create_colormap(
np.array([[0, 0, 0], [3, 3, 3], [5, 5, 5], [7, 7, 7],
[8, 8, 8]]))
] #B/W
return default_config
# INPUT INFO
experiment_id = 171102
dataset_id = 7
n_max_images = 300
# OUTPUT INFO
output_image_folder = './images_colored/'
if not os.path.exists(output_image_folder):
os.makedirs(output_image_folder)
img_filename_template = output_image_folder + 'pattern_{:06d}_{}'
img_suffix = '.png'
colored = True
colormap = create_colormap(
np.array([[255, 255, 255], [119, 255, 255], [23, 223, 252], [0, 190, 250],
[0, 158, 249], [0, 142, 249], [81, 125, 248], [150, 109, 248],
[192, 77, 247], [232, 47, 247], [255, 9, 247], [200, 0, 84]]) /
255 * 8)
# LOAD MODEL
model_path = '../../experiments/experiment_{:06d}/training/models/best_weight_model.pth'.format(
experiment_id)
#model_path = '../../prior_data/pretrained_representation/representation_000118/best_weight_model.pth'
#model_path = '../../experiments/experiment_{:06d}/repetition_{:06d}/trained_representation/saved_models/stage_{:06d}_weight_model.pth'.format(experiment_id, repetition_id, stage_id)
print("Loading the trained model ... \n")
if os.path.exists(model_path):
saved_model = torch.load(model_path, map_location='cpu')
model_cls = getattr(ad.representations.static.pytorchnnrepresentation,
saved_model['type'])
if 'self' in saved_model['init_params']:
del saved_model['init_params']['self']
def plot_holmes_partitioning(experiment_definition,
repetition_id=0,
experiment_statistics=None,
data_filters=None,
config=None,
**kwargs):
plotly_default_colors = plotly.colors.DEFAULT_PLOTLY_COLORS
plotly_default_colors_darker = [
lighten_color([
float(cc) / 255.0
for cc in c.replace("rgb(", "").replace(")", "").split(", ")
], 1.4) for c in plotly.colors.DEFAULT_PLOTLY_COLORS
]
plotly_default_colors_darker = [
"rgb({},{},{})".format(int(c[0] * 255), int(c[1] * 255),
int(c[2] * 255))
for c in plotly_default_colors_darker
]
default_config = dict(
random_seed=0,
# global style config
global_layout=dict(
height=700,
width=700,
margin=dict(l=0, r=0, b=0, t=0),
xaxis=dict(autorange=True,
showgrid=False,
zeroline=False,
showline=False,
ticks='',
showticklabels=False),
yaxis=dict(autorange=True,
showgrid=False,
zeroline=False,
showline=False,
ticks='',
showticklabels=False),
title=dict(text='',
font=dict(color="black",
size=22,
family='Times New Roman')),
hovermode='closest',
showlegend=True,
),
# Shapes style config
shapes=dict(line=dict(width=2), layer="below"),
shapes_background_colors=plotly_default_colors,
shapes_lines_colors=plotly_default_colors_darker,
# Images style config
margin_x=1,
margin_y=1,
images=dict(xref="x",
yref="y",
sizex=10,
sizey=10,
opacity=1,
xanchor="center",
yanchor="middle",
layer="above"),
images_transform=True,
images_transform_colormaps=[
create_colormap(
np.array([[255, 255, 255], [119, 255, 255], [23, 223, 252],
[0, 190, 250], [0, 158, 249], [0, 142, 249],
[81, 125, 248], [150, 109, 248], [192, 77, 247],
[232, 47, 247], [255, 9, 247], [200, 0, 84]]) / 255 *
8),
# WBPR
create_colormap(
np.array([[0, 0, 4], [0, 0, 8], [0, 4, 8], [0, 8,
8], [4, 8, 4],
[8, 8, 0], [8, 4, 0], [8, 0, 0], [4, 0, 0]])),
# BCYR
create_colormap(
np.array([[0, 2, 0], [0, 4, 0], [4, 6, 0], [8, 8,
0], [8, 4, 4],
[8, 0, 8], [4, 0, 8], [0, 0, 8], [0, 0, 4]])),
# GYPB
create_colormap(
np.array([[4, 0, 2], [8, 0, 4], [8, 0, 6], [8, 0,
8], [4, 4, 4],
[0, 8, 0], [0, 6, 0], [0, 4, 0], [0, 2, 0]])),
# PPGG
create_colormap(
np.array([[4, 4, 6], [2, 2, 4], [2, 4, 2], [4, 6, 4],
[6, 6, 4], [4, 2, 2]])), # BGYR
create_colormap(
np.array([[4, 6, 4], [2, 4, 2], [4, 4, 2], [6, 6, 4],
[6, 4, 6], [2, 2, 4]])), # GYPB
create_colormap(
np.array([[6, 6, 4], [4, 4, 2], [4, 2, 4], [6, 4, 6],
[4, 6, 6], [2, 4, 2]])), # YPCG
create_colormap(np.array([[8, 8, 8], [7, 7, 7], [5, 5, 5],
[3, 3, 3], [0, 0, 0]]),
is_marker_w=False),
# W/B
create_colormap(
np.array([[0, 0, 0], [3, 3, 3], [5, 5, 5], [7, 7, 7],
[8, 8, 8]]))
], # B/W
# Annotations style config
annotations=dict(font=dict(color="#140054",
size=18,
family='Times New Roman'),
showarrow=False,
opacity=1.0,
bgcolor='rgb(255,255,255)'))
config = ad.config.set_default_config(kwargs, config, default_config)
random.seed(config.random_seed)
np.random.seed(config.random_seed)
if repetition_id is None:
path = experiment_definition['directory']
else:
path = os.path.join(experiment_definition['directory'],
'repetition_{:06d}'.format(int(repetition_id)))
path = os.path.join(path, 'statistics')
final_observation_path = os.path.join(path, 'final_observation')
# Recover images (from png or zip)
all_images = []
if os.path.isdir(final_observation_path):
dir_content = os.listdir(final_observation_path)
for image_file_name in dir_content:
file_path = os.path.join(path, image_file_name)
if os.path.isfile(file_path) and '.png' in image_file_name:
item_id = image_file_name.split('.')[0]
item_id = int(item_id)
file = open(file_path, "rb")
image_PIL = Image.open(file)
if config.images_transform:
image_PIL = transform_image_from_colormap(
image_PIL, config.images_transform_colormaps[0])
all_images.append(image_PIL)
elif zipfile.is_zipfile(final_observation_path + '.zip'):
zf = zipfile.ZipFile(final_observation_path + '.zip', 'r')
dir_content = zf.namelist()
for image_file_name in dir_content:
item_id = image_file_name.split('.')[0]
item_id = int(item_id)
with zf.open(image_file_name) as file:
image_PIL = Image.open(file)
if config.images_transform:
image_PIL = transform_image_from_colormap(
image_PIL, config.images_transform_colormaps[0])
all_images.append(image_PIL)
else:
warnings.warn(
'No Images found under {!r}'.format(final_observation_path))
# loads pathes taken in HOLMES
ids_per_node_path = os.path.join(path, 'ids_per_node')
ids_per_node = dict(np.load(ids_per_node_path + '.npz'))
for gs_path, gs_val in ids_per_node.items():
if len(gs_val.shape) == 0:
ids_per_node[gs_path] = gs_val.dtype.type(gs_val)
pathes = []
max_depth = 0
for k in ids_per_node.keys():
cur_node_path = k[3:]
pathes.append(cur_node_path)
if (len(cur_node_path) - 1) > max_depth:
max_depth = len(cur_node_path) - 1
# load representations per node in HOLMES
representations_per_node_path = os.path.join(path, 'representations')
representations_per_node = dict(
np.load(representations_per_node_path + '.npz'))
for gs_path, gs_val in representations_per_node.items():
representations_per_node[gs_path] = gs_val.dtype.type(gs_val)
tot_width = config.global_layout.width
row_height = config.global_layout.height
figures = []
counter = 0
n_total = len(ids_per_node["gs_0"])
for depth in range(max_depth + 1):
cur_depth_pathes = []
for path in pathes:
if (len(path) - 1) == depth:
cur_depth_pathes.append(path)
cur_depth_pathes.sort()
values = []
for path in cur_depth_pathes:
n_cur_node = len(ids_per_node["gs_{}".format(path)])
values.append(n_cur_node / n_total * tot_width)
for col_idx in range(len(values)):
cur_path = cur_depth_pathes[col_idx]
value = values[col_idx]
x0 = 0
y0 = 0
# v1:
w = value
# v2:
# w = tot_width
h = row_height
# images layout
min_n_cols = 2
space_between_images = 0.1
n_cols = int((w - 2 * config.margin_x) // config.images.sizex)
n_cols = max(n_cols, min_n_cols)
space_x = config.images.sizex + space_between_images
centers_x = []
for j in range(n_cols):
centers_x.append(j * space_x + space_x / 2.0)
n_rows = int((h - 2 * config.margin_y) // config.images.sizey)
space_y = config.images.sizey + space_between_images
centers_y = []
for i in range(n_rows):
centers_y.append(i * space_y + space_y / 2.0)
cur_path_x_ids = ids_per_node["gs_{}".format(cur_path)]
indices_diverse = get_diversity_representants(
representations_per_node["gs_{}".format(cur_path)],
n_candidates=750,
n_images=n_rows * n_cols)
indices_diverse = cur_path_x_ids[indices_diverse]
#list_of_random_items = np.random.choice(cur_path_x_ids, size=n_rows * n_cols, replace=True)
images = []
for i in range(n_rows):
for j in range(n_cols):
image_config = ad.config.set_default_config(
dict(
source=all_images[indices_diverse[
np.ravel_multi_index((i, j),
dims=(n_rows, n_cols),
order='F')]],
x=x0 + config.margin_x + centers_x[j],
y=y0 + config.margin_y + centers_y[i],
), config.images)
images.append(image_config)
shape_w = n_cols * config.images.sizex + (
n_cols - 1) * space_between_images + 2 * config.margin_x
shape_h = n_rows * config.images.sizey + (
n_rows - 1) * space_between_images + 2 * config.margin_y
annotation_w = 170
annotation_h = 20
final_w = max(shape_w, annotation_w)
final_h = shape_h + annotation_h + 0.2
# annotations layout
annotation_config = ad.config.set_default_config(
dict(x=x0 + (final_w / 2),
y=y0 + shape_h + 0.2 + annotation_h,
width=annotation_w,
height=annotation_h,
text="<b>GS {}: {:.1f}%<b>".format(
cur_path, w / tot_width * 100),
font=dict(color=config.shapes_lines_colors[counter], )),
config.annotations)
# shapes layout
shape_config = ad.config.set_default_config(
dict(type='rect',
x0=x0,
y0=y0,
x1=x0 + shape_w,
y1=y0 + shape_h,
fillcolor=config.shapes_background_colors[counter],
line=dict(color=config.shapes_lines_colors[counter])),
config.shapes)
# figure
cur_fig_layout = ad.config.set_default_config(
dict(width=final_w,
height=final_h,
margin=dict(autoexpand=True),
annotations=[annotation_config],
shapes=[shape_config],
images=images), config.global_layout)
figure = dict(data=[go.Scatter()], layout=cur_fig_layout)
plotly.offline.iplot(figure)
figures.append(figure)
counter = counter + 1
if counter >= len(config.shapes_background_colors):
counter = 0
return figures