Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

378 update button for execution #396

Merged
merged 13 commits into from
Dec 7, 2022
Merged

378 update button for execution #396

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
714b341
add a buttion for refreshing
Nov 15, 2022
19563a4
using update button to trigger callback
Nov 15, 2022
2148778
Adjusted layout and fixed dianna model running error
cpranav93 Nov 18, 2022
44eefeb
Merged main after PR from #352
cpranav93 Nov 29, 2022
7dec865
Fixed the raising of missing image/model/method when not selected issue
cpranav93 Nov 29, 2022
5b714d4
silent linter
Nov 29, 2022
2d84f34
Revert "silent linter"
Nov 29, 2022
78edac0
fix linter
Nov 29, 2022
e639a04
fix pylint
Nov 29, 2022
984c427
Changed error message location
cpranav93 Dec 6, 2022
ba0c0fe
Changed spacing between button and method setting
cpranav93 Dec 7, 2022
6d1c3f8
Added a stop button
cpranav93 Dec 7, 2022
a5acbee
Merge branch 'main' into 378-update-button-for-execution
cpranav93 Dec 7, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
268 changes: 136 additions & 132 deletions dashboard/callbacks.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,147 +210,151 @@ def compute_value_i(method_sel, fn_m, fn_i):
dash.dependencies.Output('graph_img', 'figure'),
dash.dependencies.State("upload-model-img", "filename"),
dash.dependencies.State("upload-image", "filename"),
dash.dependencies.Input("signal_image", "data"),
dash.dependencies.Input("upload-model-img", "filename"),
dash.dependencies.Input("upload-image", "filename"),
dash.dependencies.Input("show_top", "value"),
dash.dependencies.Input("n_masks", "value"),
dash.dependencies.Input("feature_res", "value"),
dash.dependencies.Input("p_keep", "value"),
dash.dependencies.Input("n_samples", "value"),
dash.dependencies.Input("background", "value"),
dash.dependencies.Input("n_segments", "value"),
dash.dependencies.Input("sigma", "value"),
dash.dependencies.Input("random_state", "value")
dash.dependencies.State("signal_image", "data"),
dash.dependencies.State("upload-model-img", "filename"),
dash.dependencies.State("upload-image", "filename"),
dash.dependencies.State("show_top", "value"),
dash.dependencies.State("n_masks", "value"),
dash.dependencies.State("feature_res", "value"),
dash.dependencies.State("p_keep", "value"),
dash.dependencies.State("n_samples", "value"),
dash.dependencies.State("background", "value"),
dash.dependencies.State("n_segments", "value"),
dash.dependencies.State("sigma", "value"),
dash.dependencies.State("random_state", "value"),
dash.dependencies.Input("update_button", "n_clicks"),
dash.dependencies.Input("stop_button", "n_clicks")
)
# pylint: disable=too-many-locals
# pylint: disable=unused-argument
# pylint: disable=too-many-arguments
def update_multi_options_i(fn_m, fn_i, sel_methods, new_model, new_image,
show_top=2, n_masks=1000, feature_res=6, p_keep=0.1, n_samples=1000,
background=0, n_segments=200, sigma=0, random_state=2):
def update_multi_options_i(fn_m, fn_i, sel_methods, new_model, new_image, show_top=2, n_masks=1000, feature_res=6, p_keep=0.1, n_samples=1000,
background=0, n_segments=200, sigma=0, random_state=2, update_button=0, stop_button=0):
"""Takes in the last model and image uploaded filenames, the selected XAI method, and returns the selected XAI method."""
ctx = dash.callback_context

if ((ctx.triggered[0]["prop_id"] == "upload-model-img.filename") or
(ctx.triggered[0]["prop_id"] == "upload-image.filename") or
(not ctx.triggered)):
cache.clear()
return html.Div(['']), utilities.blank_fig()
if (not sel_methods):
return html.Div(['']), utilities.blank_fig()

# if ((ctx.triggered[0]["prop_id"] == "upload-model-img.filename") or
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If this part is no longer needed, it can maybe be deleted?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this part relates to the global cache in #418. So I do not want to delete it yet.

# (ctx.triggered[0]["prop_id"] == "upload-image.filename") or
# (not ctx.triggered)):
# cache.clear()
# return html.Div(['']), utilities.blank_fig()
# if (not sel_methods):
# return html.Div(['']), utilities.blank_fig()
if (ctx.triggered[0]["prop_id"] == "stop_button.n_clicks"):
return (html.Div(['Explanation stopped.'], style={'margin-top' : '60px'}),
utilities.blank_fig())
# update graph
if (fn_m and fn_i) is not None:
data_path = os.path.join(folder_on_server, fn_i[0])
X_test, _ = utilities.open_image(data_path)

onnx_model_path = os.path.join(folder_on_server, fn_m[0])
onnx_model = onnx.load(onnx_model_path)
# get the output node
output_node = prepare(onnx_model, gen_tensor_dict=True).outputs[0]

try:
predictions = (prepare(onnx_model).run(X_test[None, ...])
[f'{output_node}'])
if len(predictions[0]) == 2:
class_name = class_name_mnist
else:
class_name = class_names_imagenet
# get the predicted class
preds = np.array(predictions[0])
pred_class = class_name[np.argmax(preds)]
# get the top most likely results
if show_top > len(class_name):
show_top = len(class_name)
# make sure the top results are ordered most to least likely
ind = np.array(np.argpartition(preds, -show_top)[-show_top:])
ind = ind[np.argsort(preds[ind])]
ind = np.flip(ind)
top = [class_name[i] for i in ind]
n_rows = len(top)
fig = make_subplots(rows=n_rows, cols=3,
subplot_titles=("RISE", "KernelShap", "LIME"), row_titles=top,
shared_xaxes=True, vertical_spacing=0.02,
horizontal_spacing = 0.02)
# check which axis is color channel
if X_test.shape[2] <=3:
z_rise = X_test[:, :, 0]
axis_labels = {2: 'channels'}
colorscale='Bluered'
else:
z_rise = X_test[1, :, :]
axis_labels = {0: 'channels'}
colorscale='jet'
for m in sel_methods:
for i in range(n_rows):
if m == "RISE":
# RISE plot
relevances_rise = global_store_i('RISE',
onnx_model_path, X_test, labels=[ind[i]],
axis_labels=axis_labels, n_masks=n_masks,
feature_res=feature_res, p_keep=p_keep)
fig.add_trace(
go.Heatmap(z=z_rise, colorscale='gray',
showscale=False), i+1, 1)
fig.add_trace(
go.Heatmap(z=relevances_rise[0],
colorscale=colorscale, showscale=False,
opacity=0.7), i+1, 1)
elif m == "KernelSHAP":
shap_values, segments_slic = global_store_i(
m, onnx_model_path, X_test, labels=[ind[i]],
axis_labels=axis_labels, n_samples=n_samples,
background=background, n_segments=n_segments,
sigma=sigma)

# KernelSHAP plot
fig.add_trace(
go.Heatmap(z=z_rise, colorscale='gray',
showscale=False), i+1, 2)
fig.add_trace(
go.Heatmap(
z=utilities.fill_segmentation(shap_values[i][0],
segments_slic), colorscale='Bluered',
showscale=False, opacity=0.7), i+1, 2)
else:
relevances_lime = global_store_i(
m, onnx_model_path, X_test, labels=[ind[i]],
axis_labels=axis_labels, random_state=random_state)
# LIME plot
fig.add_trace(
go.Heatmap(z=z_rise, colorscale='gray',
showscale=False), i+1, 3)
fig.add_trace(
go.Heatmap(z=relevances_lime[0],
colorscale='bluered', showscale=False,
opacity=0.7), i+1, 3)

fig.update_layout(
width=650,
height=(200*n_rows+50),
paper_bgcolor=layouts.colors['blue4'])

fig.update_xaxes(showgrid=False, showticklabels=False,
zeroline=False)
fig.update_yaxes(showgrid=False, showticklabels=False,
zeroline=False, autorange="reversed")

return html.Div(['The predicted class is: ' + pred_class], style={
'fontSize': 18,
'font-weight': 'bold',
'text-decoration': 'underline',
'margin-top': '60px',
'textAlign' : 'center'
}), fig
elif (ctx.triggered[0]["prop_id"] == "update_button.n_clicks"):
if (fn_m and fn_i and sel_methods) is not None:

data_path = os.path.join(folder_on_server, fn_i[0])
X_test, _ = utilities.open_image(data_path)

onnx_model_path = os.path.join(folder_on_server, fn_m[0])
onnx_model = onnx.load(onnx_model_path)
# get the output node
output_node = prepare(onnx_model, gen_tensor_dict=True).outputs[0]

try:
predictions = (prepare(onnx_model).run(X_test[None, ...])
[f'{output_node}'])
if len(predictions[0]) == 2:
class_name = class_name_mnist
else:
class_name = class_names_imagenet
# get the predicted class
preds = np.array(predictions[0])
pred_class = class_name[np.argmax(preds)]
# get the top most likely results
if show_top > len(class_name):
show_top = len(class_name)
# make sure the top results are ordered most to least likely
ind = np.array(np.argpartition(preds, -show_top)[-show_top:])
ind = ind[np.argsort(preds[ind])]
ind = np.flip(ind)
top = [class_name[i] for i in ind]
n_rows = len(top)
fig = make_subplots(rows=n_rows, cols=3,
subplot_titles=("RISE", "KernelShap", "LIME"), row_titles=top,
shared_xaxes=True, vertical_spacing=0.02,
horizontal_spacing = 0.02)
# check which axis is color channel
if X_test.shape[2] <=3:
z_rise = X_test[:, :, 0]
axis_labels = {2: 'channels'}
colorscale='Bluered'
else:
z_rise = X_test[1, :, :]
axis_labels = {0: 'channels'}
colorscale='jet'
for m in sel_methods:
for i in range(n_rows):
if m == "RISE":
# RISE plot
relevances_rise = global_store_i('RISE',
onnx_model_path, X_test, labels=[ind[i]],
axis_labels=axis_labels, n_masks=n_masks,
feature_res=feature_res, p_keep=p_keep)
fig.add_trace(
go.Heatmap(z=z_rise, colorscale='gray',
showscale=False), i+1, 1)
fig.add_trace(
go.Heatmap(z=relevances_rise[0],
colorscale=colorscale, showscale=False,
opacity=0.7), i+1, 1)
elif m == "KernelSHAP":
shap_values, segments_slic = global_store_i(
m, onnx_model_path, X_test, labels=[ind[i]],
axis_labels=axis_labels, n_samples=n_samples,
background=background, n_segments=n_segments,
sigma=sigma)

# KernelSHAP plot
fig.add_trace(
go.Heatmap(z=z_rise, colorscale='gray',
showscale=False), i+1, 2)
fig.add_trace(
go.Heatmap(
z=utilities.fill_segmentation(shap_values[i][0],
segments_slic), colorscale='Bluered',
showscale=False, opacity=0.7), i+1, 2)
else:
relevances_lime = global_store_i(
m, onnx_model_path, X_test, labels=[ind[i]],
axis_labels=axis_labels, random_state=random_state)
# LIME plot
fig.add_trace(
go.Heatmap(z=z_rise, colorscale='gray',
showscale=False), i+1, 3)
fig.add_trace(
go.Heatmap(z=relevances_lime[0],
colorscale='bluered', showscale=False,
opacity=0.7), i+1, 3)
fig.update_layout(
width=650,
height=(200*n_rows+50),
paper_bgcolor=layouts.colors['blue4'])

except Exception as e:
print(e)
return (html.Div(['There was an error running the model. Check' +
'either the test image or the model.']), utilities.blank_fig())
else:
return (html.Div(['Missing either model or image.']),
utilities.blank_fig())
fig.update_xaxes(showgrid=False, showticklabels=False,
zeroline=False)
fig.update_yaxes(showgrid=False, showticklabels=False,
zeroline=False, autorange="reversed")

return html.Div(['The predicted class is: ' + pred_class], style={
'fontSize': 18,
'font-weight': 'bold',
'text-decoration': 'underline',
'margin-top': '60px',
'textAlign' : 'center'
}), fig

except Exception as e:
print(e)
return (html.Div(['There was an error running the model. Check' +
'either the test image or the model.']), utilities.blank_fig())
else:
return (html.Div(['Missing model, image or XAI method.'], style={'margin-top' : '60px'}),
utilities.blank_fig())

###################################################################

Expand Down
41 changes: 39 additions & 2 deletions dashboard/layouts.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
'blue2' : '#0e749b',
'blue3' : '#15b3f0',
'blue4' : '#E4F3F9', #light blue
'red1' : '#FF0000',
'yellow1' : '#f0d515'
}

Expand Down Expand Up @@ -290,13 +291,49 @@ def get_uploads_images():
], className = 'three columns'
),
],
className = 'row', style = {'padding-bottom' : '3%'}
className = 'row', style = {'padding-bottom' : '1%'}
),
html.Div([
# update button
html.Div([
html.Button('Update explanation',
id='update_button',
n_clicks=0,
style={
'margin-left': '0px',
'margin-top': '0px',
'width': '20%',
'float': 'left',
'backgroundColor': colors['blue2'],
'color' : colors['white']
}
),
],
),
html.Div([
html.Button('Stop Explanation',
id='stop_button',
n_clicks=0,
style={
'margin-left': '40px',
'margin-top': '0px',
'width': '20%',
'float': 'left',
'backgroundColor': colors['red1'],
'color' : colors['white']
}
),
],
),
],
className = 'row', style = {'padding-bottom' : '1%'}
),
# Settings bar
html.Div([
html.Div([
html.H6(children='XAI method specific settings',
style={'font-weight': 'bold'}),
style={'font-weight': 'bold',
'margin-top': '30px',}),
], className='nine columns'
),
],
Expand Down