steffen/self-replicating-neural-networks
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Visual Debugging

Browse Source
This commit is contained in:
Si11ium 2019-03-05 20:31:33 +01:00
parent 6625481091
commit de6aa68f23
6 changed files with 59 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
code/experiment.py
View File

@ -67,7 +67,7 @@ class FixpointExperiment(Experiment):
net.self_attack()
i += 1
if run_id:
weights = net.get_weights()
weights = net.get_weights_flat()
self.add_trajectory_segment(run_id, weights)
self.count(net)

7
code/latent_trajectory_plot.html Normal file
View File

File diff suppressed because one or more lines are too long

1
code/methods.py
View File

@ -173,6 +173,7 @@ class FeedForwardNetwork(_BaseNetwork):
bar.update()
return losses
if __name__ == '__main__':
with Experiment() as exp:
features, cells, layers = 2, 2, 2

7
code/network.py
View File

@ -83,6 +83,9 @@ class NeuralNetwork(PrintingObject):
def get_weights(self):
return self.model.get_weights()
def get_weights_flat(self):
return np.hstack([weight.flatten() for weight in self.get_weights()])
def set_weights(self, new_weights):
return self.model.set_weights(new_weights)
@ -603,9 +606,9 @@ class TrainingNeuralNetworkDecorator(NeuralNetwork):
if __name__ == '__main__':
if False:
if True:
with FixpointExperiment() as exp:
for run_id in tqdm(range(1)):
for run_id in tqdm(range(100)):
# net = WeightwiseNeuralNetwork(width=2, depth=2).with_keras_params(activation='linear')
# net = AggregatingNeuralNetwork(aggregates=4, width=2, depth=2)\
net = FFTNeuralNetwork(aggregates=4, width=2, depth=2) \

7
code/plot.html Normal file
View File

File diff suppressed because one or more lines are too long

77
code/visualization.py
View File

@ -22,47 +22,42 @@ def build_args():
def plot_latent_trajectories(data_dict, filename='latent_trajectory_plot'):
# TODO Fist and Last Position Markers
def norm(val, a=0, b=0.25):
return (val - a) / (b - a)
bupu = cl.scales['9']['seq']['BuPu']
scale = cl.interp(bupu, len(data_dict)) # Map color scale to N bins
scale = cl.interp(bupu, len(data_dict)+1) # Map color scale to N bins
# Fit the mebedding space
transformer = TSNE()
for trajectory in data_dict:
transformer.fit(trajectory)
for trajectory_id in data_dict:
transformer.fit(np.asarray(data_dict[trajectory_id]))
# Transform data accordingly and plot it
data = []
for t_id, trajectory in enumerate(data_dict):
transformed = transformer.fit(trajectory)
for trajectory_id in data_dict:
transformed = transformer._fit(np.asarray(data_dict[trajectory_id]))
line_trace = go.Scatter(
x=transformed[:, 0],
y=transformed[:, 1],
text='Hovertext goes here'.format(),
line=dict(color=scale[t_id]),
line=dict(color=scale[trajectory_id]),
# legendgroup='Position -{}'.format(pos),
# name='Position -{}'.format(pos),
showlegend=False,
# hoverinfo='text',
mode='lines')
line_start = go.Scatter(mode='markers', x=transformed[0, 0], y=transformed[0, 1],
line_start = go.Scatter(mode='markers', x=[transformed[0, 0]], y=[transformed[0, 1]],
marker=dict(
color='rgb(255, 0, 0)',
size=2
size=4
),
showlegend=False
)
line_end = go.Scatter(mode='markers', x=transformed[-1, 0], y=transformed[-1, 1],
line_end = go.Scatter(mode='markers', x=[transformed[-1, 0]], y=[transformed[-1, 1]],
marker=dict(
color='rgb(0, 0, 0)',
size=2
size=4
),
showlegend=False
)
)
data.extend([line_trace, line_start, line_end])
layout = dict(title='{} - Latent Trajectory Movement'.format('Penis'),
@ -74,30 +69,30 @@ def plot_latent_trajectories(data_dict, filename='latent_trajectory_plot'):
pass
def plot_latent_trajectories_3D(param_dict, filename='plot'):
def plot_latent_trajectories_3D(data_dict, filename='plot'):
def norm(val, a=0, b=0.25):
return (val - a) / (b - a)
bupu = cl.scales['9']['seq']['BuPu']
scale = cl.interp(bupu, len(param_dict.get('trajectories', []))) # Map color scale to N bins
scale = cl.interp(bupu, len(data_dict)+1) # Map color scale to N bins
max_len = max([len(trajectory) for trajectory in param_dict.get('trajectories', [])])
max_len = max([len(trajectory) for trajectory in data_dict.values()])
# Fit the mebedding space
transformer = TSNE()
for trajectory in param_dict.get('trajectories', []):
transformer.fit(trajectory)
for trajectory_id in data_dict:
transformer.fit(data_dict[trajectory_id])
# Transform data accordingly and plot it
data = []
for t_id, trajectory in enumerate(param_dict.get('trajectories', [])):
transformed = transformer.fit(trajectory)
for trajectory_id in data_dict:
transformed = transformer._fit(np.asarray(data_dict[trajectory_id]))
trace = go.Scatter3d(
x=transformed[:, 0],
y=transformed[:, 1],
z=np.arange(max(max_len)),
z=np.arange(transformed.shape[0]),
text='Hovertext goes here'.format(),
line=dict(color=scale[t_id]),
line=dict(color=scale[trajectory_id]),
# legendgroup='Position -{}'.format(pos),
# name='Position -{}'.format(pos),
showlegend=False,
@ -106,9 +101,9 @@ def plot_latent_trajectories_3D(param_dict, filename='plot'):
data.append(trace)
layout = go.Layout(scene=dict(aspectratio=dict(x=2, y=2, z=1),
xaxis=dict(tickwidth=1, title='Number of Cells'),
yaxis=dict(tickwidth=1, title='Number of Layers'),
zaxis=dict(tickwidth=1, title='Position -pX')),
xaxis=dict(tickwidth=1, title='Transformed X'),
yaxis=dict(tickwidth=1, title='transformed Y'),
zaxis=dict(tickwidth=1, title='Epoch')),
title='{} - Latent Trajectory Movement'.format('Penis'),
width=800, height=800,
margin=dict(l=0, r=0, b=0, t=0))
@ -125,18 +120,18 @@ def plot_histogram(bars_dict_list, filename='histogram_plot'):
data = []
for bar_id, bars_dict in bars_dict_list:
hist = go.Histogram(
histfunc="count",
y=bars_dict.get('value', 14),
x=bars_dict.get('name', 'gimme a name'),
showlegend=False,
marker=dict(
color=ryb[bar_id]
),
)
histfunc="count",
y=bars_dict.get('value', 14),
x=bars_dict.get('name', 'gimme a name'),
showlegend=False,
marker=dict(
color=ryb[bar_id]
),
)
data.append(hist)
layout=dict(title='{} Histogram Plot'.format('Experiment Name Penis'),
height=400, width=400, margin=dict(l=0, r=0, t=0, b=0))
height=400, width=400, margin=dict(l=0, r=0, t=0, b=0))
fig = go.Figure(data=data, layout=layout)
pl.offline.plot(fig, auto_open=True, filename=filename)
@ -161,7 +156,7 @@ def line_plot(line_dict_list, filename='lineplot'):
marker=dict(color="#444"),
line=dict(width=0),
fillcolor=rdylgn_background[line_id],
)
)
trace = go.Scatter(
x=line_dict['x'],
@ -183,7 +178,7 @@ def line_plot(line_dict_list, filename='lineplot'):
data.extend([upper_bound, trace, lower_bound])
layout=dict(title='{} Line Plot'.format('Experiment Name Penis'),
height=800, width=800, margin=dict(l=0, r=0, t=0, b=0))
height=800, width=800, margin=dict(l=0, r=0, t=0, b=0))
fig = go.Figure(data=data, layout=layout)
pl.offline.plot(fig, auto_open=True, filename=filename)
@ -195,4 +190,8 @@ if __name__ == '__main__':
in_file = args.in_file[0]
out_file = args.out_file
with open(in_file, 'rb') as in_f:
experiment = dill.load(in_f)
plot_latent_trajectories_3D(experiment.data_storage)
print('aha')