def register(self, modules):
# here modules are assumed to be a dictionary
for key in modules:
self.keys.append(key)
self.values[key] = {}
self.values[key]['dtype'] = modules[key]['dtype']
self.values[key]['vtype'] = modules[key]['vtype']
self.values[key]['win'] = modules[key]['win'] \
if 'win' in modules[key].keys() \
else None
if modules[key]['vtype'] == 'plot':
self.values[key]['layout'] = modules[key]['layout'] \
if 'layout' in modules[key].keys() \
else {'windows': [key], 'id': 0}
self.values[key]['value'] = []
# Create traces
data = [
go.Scatter(x=[],
y=[],
mode='lines',
name=self.values[key]['layout']['windows'][i])
for i in range(len(self.values[key]['layout']['windows']))
]
# Edit the layout
layout = dict(
title=key,
xaxis=dict(title='Epoch'),
yaxis=dict(title=key),
)
fig = go.Figure(data=data, layout=layout)
fig = fig.to_dict()
self.values[key]['win'] = self.viz._send(
plotlify(fig, env=self.env, win=self.values[key]['win']))
elif modules[key]['vtype'] == 'image':
self.values[key]['value'] = None
elif modules[key]['vtype'] == 'images':
self.values[key]['value'] = None
else:
raise Exception('Data type not supported, please update the '
'visualizer plugin and rerun !!')
def update(self, modules):
for key in modules:
if self.values[key]['dtype'] == 'scalar':
self.values[key]['value'].append(modules[key])
elif self.values[key]['dtype'] == 'image':
self.values[key]['value'] = modules[key]
elif self.values[key]['dtype'] == 'images':
self.values[key]['value'] = modules[key]
else:
raise Exception('Data type not supported, please update the '
'visualizer plugin and rerun !!')
for key in self.keys:
if self.values[key]['vtype'] == 'plot':
# Create traces
data = [
go.Scatter(
x=[],
y=[],
)
for i in range(len(self.values[key]['layout']['windows']))
]
data[self.values[key]['layout']['id']] = go.Scatter(
x=np.array([self.iteration]).tolist(),
y=np.array([self.values[key]['value'][-1]]).tolist(),
)
fig = dict(data=data, append=True)
self.viz._send(plotlify(fig,
env=self.env,
win=self.values[key]['win']),
endpoint='update')
elif self.values[key]['vtype'] == 'image':
temp = self.values[key]['value'].numpy()
for i in range(temp.shape[0]):
temp[i] = temp[i] - temp[i].min()
temp[i] = temp[i] / temp[i].max()
if self.iteration == 0:
self.values[key]['win'] = self.viz.image(
temp,
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
else:
self.viz.image(temp,
win=self.values[key]['win'],
opts=dict(title=key,
caption=self.iteration))
elif self.values[key]['vtype'] == 'images':
temp = self.values[key]['value'].numpy()
for i in range(temp.shape[0]):
for j in range(temp.shape[1]):
temp[i][j] = temp[i][j] - temp[i][j].min()
temp[i][j] = temp[i][j] / temp[i][j].max()
if self.iteration == 0:
self.values[key]['win'] = self.viz.images(
temp,
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
else:
self.viz.images(temp,
opts=dict(title=key,
caption=self.iteration),
win=self.values[key]['win'])
else:
raise Exception('Visualization type not supported, please '
'update the visualizer plugin and rerun !!')
self.iteration = self.iteration + 1
def update(self, modules):
for key in modules:
if self.values[key]['dtype'] == 'scalar':
self.values[key]['value'].append(modules[key])
elif self.values[key]['dtype'] == 'vector':
if len(self.values[key]['value']) == 0:
self.values[key]['value'] = modules[key].numpy()
else:
self.values[key]['value'] = np.concatenate(
(self.values[key]['value'], modules[key].numpy()),
axis=0)
elif self.values[key]['dtype'] in ('image', 'images', 'heatmap'):
self.values[key]['value'] = modules[key]
elif self.values[key]['dtype'] in ('heatmaps', ):
input, label, output = modules[key]
images = list()
# output_landmarks = self.toLandmarks(output)
for i in range(len(input)):
images.append(input[i])
images.append(0.5 * input[i] + 0.5 *
self.toColorHeatmap(label[i].sum(
dim=0), input[i].size()[1:]))
for hm_i in range(len(output[-1].cpu().data[i])):
images.append(
0.5 * input[i] + 0.5 *
self.toColorHeatmap(output[-1].cpu().data[i][hm_i],
input[i].size()[1:]))
self.values[key]['value'] = torch.stack(images)
else:
raise Exception('Data type not supported, please update the '
'visualizer plugin and rerun !!')
for key in self.keys:
if self.values[key]['vtype'] == 'plot':
# Create traces
data = [
go.Scatter(
x=[],
y=[],
)
for i in range(len(self.values[key]['layout']['windows']))
]
data[self.values[key]['layout']['id']] = go.Scatter(
x=np.array([self.iteration]).tolist(),
y=np.array([self.values[key]['value'][-1]]).tolist(),
)
fig = dict(data=data, append=True)
self.viz._send(plotlify(fig,
env=self.env,
win=self.values[key]['win']),
endpoint='update')
elif self.values[key]['vtype'] == 'scatter':
# Create traces
data = [
go.Scatter(
x=[],
y=[],
)
for i in range(len(self.values[key]['layout']['windows']))
]
values = self.values[key]['value']
data[self.values[key]['layout']['id']] = go.Scatter(
x=values[:, 0].tolist(),
y=values[:, 1].tolist(),
)
fig = dict(data=data, append=True)
self.viz._send(plotlify(fig,
env=self.env,
win=self.values[key]['win']),
endpoint='update')
elif self.values[key]['vtype'] == 'histogram':
# Create traces
# print(self.values[key]['value'].shape)
data = [
go.Histogram(
x=[],
histnorm='probability'
# name=self.values[key]['layout']['windows'][i]
)
for i in range(len(self.values[key]['layout']['windows']))
]
data[self.values[key]['layout']['id']] = go.Histogram(
x=self.values[key]['value'].numpy().tolist(),
histnorm='probability')
layout = dict(
title=key,
xaxis=dict(title='Epoch'),
)
fig = dict(data=data, layout=layout)
self.values[key]['win'] = self.viz._send(
plotlify(fig, env=self.env, win=self.values[key]['win']))
elif self.values[key]['vtype'] == 'image':
temp = self.values[key]['value'].numpy()
for i in range(temp.shape[0]):
temp[i] = temp[i] - temp[i].min()
temp[i] = temp[i] / temp[i].max()
if self.iteration == 0:
self.values[key]['win'] = self.viz.image(
temp,
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
else:
self.viz.image(temp,
win=self.values[key]['win'],
opts=dict(title=key,
caption=self.iteration))
elif self.values[key]['vtype'] == 'images':
temp = self.values[key]['value'].numpy()
for i in range(temp.shape[0]):
for j in range(temp.shape[1]):
temp[i][j] = temp[i][j] - temp[i][j].min()
temp[i][j] = temp[i][j] / temp[i][j].max()
if self.iteration == 0:
self.values[key]['win'] = self.viz.images(
temp,
nrow=11, # FIXME: needs to be automated later
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
else:
self.viz.images(
temp,
nrow=11, # FIXME: needs to be automated later
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
elif self.values[key]['vtype'] == 'heatmap':
temp = self.values[key]['value'].numpy().astype(
np.float64) #.numpy().astype(np.float64)
if self.iteration == 0:
self.values[key]['win'] = self.viz.surf(
X=temp,
win=self.values[key]['win'],
opts=dict(title=key,
colormap='Hot',
caption=self.iteration))
else:
self.viz.surf(X=temp,
win=self.values[key]['win'],
opts=dict(title=key,
colormap='Hot',
caption=self.iteration))
elif self.values[key]['vtype'] == 'heatmaps':
fig = tools.make_subplots(rows=2, cols=9)
X = np.squeeze(X)
assert X.ndim == 2, 'X should be two-dimensional'
opts = {} if opts is None else opts
opts['xmin'] = opts.get('xmin', X.min())
opts['xmax'] = opts.get('xmax', X.max())
opts['colormap'] = opts.get('colormap', 'Viridis')
_assert_opts(opts)
data = [{
'z': X.tolist(),
'cmin': opts['xmin'],
'cmax': opts['xmax'],
'type': stype,
'colorscale': opts['colormap']
}]
fig.append_trace(trace1, 1, 1)
fig.append_trace(trace2, 1, 2)
fig['layout'].update(height=600,
width=600,
title='i <3 subplots')
else:
raise Exception('Visualization type not supported, please '
'update the visualizer plugin and rerun !!')
self.iteration = self.iteration + 1
def update(self, modules):
for key in modules:
if self.values[key]['dtype'] == 'scalar':
self.values[key]['value'].append(modules[key])
elif self.values[key]['dtype'] in ('image', 'images'):
self.values[key]['value'] = modules[key]
else:
raise Exception('Data type not supported, please update the '
'visualizer plugin and rerun !!')
for key in self.keys:
if self.values[key]['vtype'] == 'plot':
# Create traces
data = [
go.Scatter(
x=[],
y=[],
)
for i in range(len(self.values[key]['layout']['windows']))
]
try:
data[self.values[key]['layout']['id']] = go.Scatter(
x=np.array([self.iteration]).tolist(),
y=np.array([self.values[key]['value'][-1]]).tolist(),
)
except Exception as e:
import pdb
pdb.set_trace() # breakpoint 344a72de //
fig = dict(data=data, append=True)
self.viz._send(plotlify(fig,
env=self.env,
win=self.values[key]['win']),
endpoint='update')
elif self.values[key]['vtype'] == 'image' and self.values[key][
'value'] is not None:
if torch.is_tensor(self.values[key]['value']):
temp = self.values[key]['value'].numpy(
) if torch.is_tensor(self.values[key]['value']
) else self.values[key]['value']
for i in range(temp.shape[0]):
temp[i] = temp[i] - temp[i].min()
temp[i] = temp[i] / temp[i].max()
else:
temp = self.values[key]['value']
if self.iteration == 0:
self.values[key]['win'] = self.viz.image(
temp,
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
else:
self.viz.image(temp,
win=self.values[key]['win'],
opts=dict(title=key,
caption=self.iteration))
elif self.values[key]['vtype'] == 'images' and self.values[key][
'value'] is not None:
if torch.is_tensor(self.values[key]['value']):
temp = self.values[key]['value'].numpy(
) if torch.is_tensor(self.values[key]['value']
) else self.values[key]['value']
for i in range(temp.shape[0]):
for j in range(temp.shape[1]):
temp[i][j] = temp[i][j] - temp[i][j].min()
temp[i][j] = temp[i][j] / temp[i][j].max()
else:
temp = self.values[key]['value']
if self.iteration == 0:
self.values[key]['win'] = self.viz.images(
temp,
nrow=11, # TODO: needs to be automated later
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
else:
self.viz.images(
temp,
nrow=11, # TODO: needs to be automated later
win=self.values[key]['win'],
opts=dict(title=key, caption=self.iteration))
else:
pass
# raise Exception('Visualization type not supported, please '
# 'update the visualizer plugin and rerun !!')
self.iteration = self.iteration + 1
def register(self, modules, reset=True):
# here modules are assumed to be a dictionary
for key in modules:
self.keys.append(key)
self.values[key] = {}
self.values[key]['dtype'] = modules[key]['dtype']
self.values[key]['vtype'] = modules[key]['vtype']
self.values[key]['win'] = modules[key]['win'] \
if 'win' in modules[key].keys() \
else None
if modules[key]['vtype'] == 'plot':
self.values[key]['layout'] = modules[key]['layout'] \
if 'layout' in modules[key].keys() \
else {'windows': [key], 'id': 0}
self.values[key]['value'] = []
# Create traces
data = [
go.Scatter(x=[],
y=[],
mode='lines',
name=self.values[key]['layout']['windows'][i])
for i in range(len(self.values[key]['layout']['windows']))
]
# Edit the layout
layout = dict(
title=key,
xaxis=dict(title='Epoch'),
yaxis=dict(title=key),
)
fig = dict(data=data, layout=layout)
if reset:
self.values[key]['win'] = self.viz._send(
plotlify(fig,
env=self.env,
win=self.values[key]['win']))
elif modules[key]['vtype'] == 'scatter':
self.values[key]['layout'] = modules[key]['layout'] \
if 'layout' in modules[key].keys() \
else {'windows': [key], 'id': 0}
self.values[key]['value'] = []
# Create traces
data = [
go.Scatter(x=[],
y=[],
mode='markers',
name=self.values[key]['layout']['windows'][i])
for i in range(len(self.values[key]['layout']['windows']))
]
# Edit the layout
layout = dict(
title=key,
xaxis=dict(title='x'),
yaxis=dict(title='y'),
)
fig = dict(data=data, layout=layout)
if reset:
self.values[key]['win'] = self.viz._send(
plotlify(fig,
env=self.env,
win=self.values[key]['win']))
elif modules[key]['vtype'] == 'histogram':
self.values[key]['value'] = np.array([])
self.values[key]['layout'] = modules[key]['layout'] \
if 'layout' in modules[key].keys() \
else {'windows': [key], 'id': 0}
self.values[key]['value'] = []
# Create traces
data = [
go.Histogram(
x=[],
histnorm='probability'
# name=self.values[key]['layout']['windows'][i]
)
for i in range(len(self.values[key]['layout']['windows']))
]
# Edit the layout
layout = dict(
title=key,
xaxis=dict(title='Epoch'),
)
fig = dict(data=data, layout=layout)
if reset:
self.values[key]['win'] = self.viz._send(
plotlify(fig,
env=self.env,
win=self.values[key]['win']))
elif modules[key]['vtype'] in ('image', 'images', 'heatmap',
'heatmaps'):
self.values[key]['value'] = None
else:
raise Exception('Data type not supported, please update the '
'visualizer plugin and rerun !!')