Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
T
Troubled Cell Detection
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Requirements
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Locked files
Build
Pipelines
Jobs
Pipeline schedules
Test cases
Artifacts
Deploy
Releases
Package registry
Container registry
Model registry
Operate
Environments
Terraform modules
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Code review analytics
Issue analytics
Insights
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
GitLab community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
Laura Christine Kühle
Troubled Cell Detection
Commits
b2e66543
Commit
b2e66543
authored
Dec 7, 2021
by
Laura Christine Kühle
Browse files
Options
Downloads
Patches
Plain Diff
Added option to compare evaluation of multiple models.
parent
9adbf4a6
Branches
Branches containing commit
No related tags found
No related merge requests found
Changes
4
Show whitespace changes
Inline
Side-by-side
Showing
4 changed files
ANN_Training.py
+61
-42
61 additions, 42 deletions
ANN_Training.py
Plotting.py
+5
-15
5 additions, 15 deletions
Plotting.py
Snakefile
+20
-39
20 additions, 39 deletions
Snakefile
config.yaml
+17
-8
17 additions, 8 deletions
config.yaml
with
103 additions
and
104 deletions
ANN_Training.py
+
61
−
42
View file @
b2e66543
...
...
@@ -2,7 +2,7 @@
"""
@author: Laura C. Kühle, Soraya Terrab (sorayaterrab)
TODO: Give option to compare multiple models
TODO: Give option to compare multiple models
-> Done
TODO: Add more evaluation measures (AUROC, ROC, F1, training accuracy, boxplot over CVF, etc.) -> Done
TODO: Add log to pipeline
TODO: Remove object set-up
...
...
@@ -11,6 +11,7 @@ TODO: Improve maximum selection runtime
TODO: Discuss if we want training accuracy/ROC in addition to CFV
TODO: Discuss whether to change output to binary
TODO: Adapt TCD file to new classification
TODO: Improve classification stat handling
"""
import
numpy
as
np
...
...
@@ -33,7 +34,7 @@ class ModelTrainer(object):
def
_reset
(
self
,
config
):
self
.
_dir
=
config
.
pop
(
'
dir
'
,
'
test_data
'
)
self
.
_model_name
=
config
.
pop
(
'
model_name
'
,
'
0
'
)
self
.
_read_training_data
()
self
.
_
training_data
=
read_training_data
(
self
.
_dir
)
self
.
_batch_size
=
config
.
pop
(
'
batch_size
'
,
min
(
len
(
self
.
_training_data
)
//
2
,
500
))
self
.
_num_epochs
=
config
.
pop
(
'
num_epochs
'
,
1000
)
...
...
@@ -63,13 +64,6 @@ class ModelTrainer(object):
self
.
_model
.
parameters
(),
**
self
.
_optimizer_config
)
self
.
_validation_loss
=
torch
.
zeros
(
self
.
_num_epochs
//
10
)
def
_read_training_data
(
self
):
# Get training dataset from saved file and map to Torch tensor and dataset
input_file
=
self
.
_dir
+
'
/input_data.npy
'
output_file
=
self
.
_dir
+
'
/output_data.npy
'
self
.
_training_data
=
TensorDataset
(
*
map
(
torch
.
tensor
,
(
np
.
load
(
input_file
),
np
.
load
(
output_file
))))
def
epoch_training
(
self
,
dataset
=
None
,
num_epochs
=
None
):
# Split data into training and validation set
if
dataset
is
None
:
...
...
@@ -108,39 +102,7 @@ class ModelTrainer(object):
if
valid_loss
/
len
(
valid_dl
)
<
self
.
_threshold
:
break
def
test_model
(
self
,
num_iterations
=
100
):
classification_stats
=
[]
for
iteration
in
range
(
num_iterations
):
dataset
=
self
.
_training_data
for
train_index
,
test_index
in
KFold
(
n_splits
=
5
,
shuffle
=
True
).
split
(
dataset
):
# print("TRAIN:", train_index, "TEST:", test_index)
training_set
=
TensorDataset
(
*
dataset
[
train_index
])
test_set
=
dataset
[
test_index
]
classification_stats
.
append
(
self
.
_test_fold
(
training_set
,
test_set
))
# print(classification_stats)
# print(np.array(classification_stats).mean(axis=0))
plot_boxplot
([
self
.
_model_name
],
*
np
.
array
(
classification_stats
).
transpose
())
classification_stats
=
np
.
array
(
classification_stats
).
mean
(
axis
=
0
)
plot_classification_accuracy
([
self
.
_model_name
],
*
classification_stats
)
# Set paths for plot files if not existing already
plot_dir
=
self
.
_dir
+
'
/model evaluation
'
if
not
os
.
path
.
exists
(
plot_dir
):
os
.
makedirs
(
plot_dir
)
# Save plots
for
identifier
in
plt
.
get_figlabels
():
# Set path for figure directory if not existing already
if
not
os
.
path
.
exists
(
plot_dir
+
'
/
'
+
identifier
):
os
.
makedirs
(
plot_dir
+
'
/
'
+
identifier
)
plt
.
figure
(
identifier
)
plt
.
savefig
(
plot_dir
+
'
/
'
+
identifier
+
'
/
'
+
self
.
_model_name
+
'
.pdf
'
)
def
_test_fold
(
self
,
training_set
,
test_set
):
def
test_model
(
self
,
training_set
,
test_set
):
self
.
epoch_training
(
training_set
,
num_epochs
=
100
)
self
.
_model
.
eval
()
...
...
@@ -185,6 +147,63 @@ class ModelTrainer(object):
pass
def
read_training_data
(
directory
):
# Get training dataset from saved file and map to Torch tensor and dataset
input_file
=
directory
+
'
/input_data.npy
'
output_file
=
directory
+
'
/output_data.npy
'
return
TensorDataset
(
*
map
(
torch
.
tensor
,
(
np
.
load
(
input_file
),
np
.
load
(
output_file
))))
def
evaluate_models
(
models
,
directory
,
num_iterations
=
100
):
dataset
=
read_training_data
(
directory
)
stats
=
[
'
Precision
'
,
'
Recall
'
,
'
Accuracy
'
,
'
F-Score
'
,
'
AUROC
'
]
classification_stats
=
{
model
:
{
name
:
[]
for
name
in
stats
}
for
model
in
models
}
for
iteration
in
range
(
num_iterations
):
for
train_index
,
test_index
in
KFold
(
n_splits
=
5
,
shuffle
=
True
).
split
(
dataset
):
# print("TRAIN:", train_index, "TEST:", test_index)
training_set
=
TensorDataset
(
*
dataset
[
train_index
])
test_set
=
dataset
[
test_index
]
for
model
in
models
:
result
=
models
[
model
].
test_model
(
training_set
,
test_set
)
count
=
0
for
stat
in
stats
:
classification_stats
[
model
][
stat
].
append
(
result
[
count
])
count
+=
1
# print(classification_stats)
# print(np.array(classification_stats).mean(axis=0))
# print(np.array(classification_stats['Adam']['Precision']).shape)
# print(np.array([np.array(classification_stats[model]) for model in models]).transpose().shape)
# print(np.array([np.array(classification_stats[model]).transpose() for model in models]).shape)
# print(np.array([[classification_stats[model][stat] for model in models] for stat in stats]).shape)
# print(np.array([[np.array(classification_stats[model][stat]).mean(axis=0) for model in models] for stat in stats]).shape)
# print(np.array([*(np.array([[classification_stats[model][stat]
# for model in models] for stat in stats]))]).shape)
# print(*(np.array([[classification_stats[model][stat]
# for model in models] for stat in stats]))[0].shape)
plot_boxplot
(
models
.
keys
(),
*
(
np
.
array
([[
classification_stats
[
model
][
stat
]
for
model
in
models
]
for
stat
in
stats
])))
classification_stats
=
[[
np
.
array
(
classification_stats
[
model
][
stat
]).
mean
(
axis
=
0
)
for
model
in
models
]
for
stat
in
stats
]
# print(*classification_stats)
plot_classification_accuracy
(
models
.
keys
(),
*
classification_stats
)
# Set paths for plot files if not existing already
plot_dir
=
directory
+
'
/model evaluation
'
if
not
os
.
path
.
exists
(
plot_dir
):
os
.
makedirs
(
plot_dir
)
# Save plots
for
identifier
in
plt
.
get_figlabels
():
# Set path for figure directory if not existing already
if
not
os
.
path
.
exists
(
plot_dir
+
'
/
'
+
identifier
):
os
.
makedirs
(
plot_dir
+
'
/
'
+
identifier
)
plt
.
figure
(
identifier
)
plt
.
savefig
(
plot_dir
+
'
/
'
+
identifier
+
'
/
'
+
'
_
'
.
join
(
models
.
keys
())
+
'
.pdf
'
)
# Loss Functions: BCELoss, BCEWithLogitsLoss,
# CrossEntropyLoss (not working), MSELoss (with reduction='sum')
# Optimizer: Adam, SGD
...
...
This diff is collapsed.
Click to expand it.
Plotting.py
+
5
−
15
View file @
b2e66543
...
...
@@ -253,11 +253,6 @@ def plot_classification_accuracy(xlabels, precision, recall, accuracy, fscore, a
List of strings for x-axis labels.
"""
precision
=
[
precision
]
recall
=
[
recall
]
accuracy
=
[
accuracy
]
fscore
=
[
fscore
]
auroc
=
[
auroc
]
pos
=
np
.
arange
(
len
(
xlabels
))
width
=
1
/
(
3
*
len
(
xlabels
))
fig
=
plt
.
figure
(
'
classification_accuracy
'
)
...
...
@@ -278,25 +273,20 @@ def plot_classification_accuracy(xlabels, precision, recall, accuracy, fscore, a
def
plot_boxplot
(
xlabels
,
precision
,
recall
,
accuracy
,
fscore
,
auroc
):
precision
=
[
precision
]
recall
=
[
recall
]
accuracy
=
[
accuracy
]
fscore
=
[
fscore
]
auroc
=
[
auroc
]
fig
=
plt
.
figure
(
'
boxplot_accuracy
'
)
pos
=
np
.
arange
(
len
(
xlabels
))
width
=
1
/
(
5
*
len
(
xlabels
))
ax
=
fig
.
add_axes
([
0.15
,
0.1
,
0.75
,
0.8
])
boxplots
=
[]
boxplots
.
append
(
ax
.
boxplot
(
fscore
,
positions
=
pos
-
3
*
width
,
widths
=
width
,
meanline
=
True
,
boxplots
.
append
(
ax
.
boxplot
(
fscore
.
transpose
()
,
positions
=
pos
-
3
*
width
,
widths
=
width
,
meanline
=
True
,
showmeans
=
True
,
patch_artist
=
True
))
boxplots
.
append
(
ax
.
boxplot
(
precision
,
positions
=
pos
-
1.5
*
width
,
widths
=
width
,
meanline
=
True
,
boxplots
.
append
(
ax
.
boxplot
(
precision
.
transpose
()
,
positions
=
pos
-
1.5
*
width
,
widths
=
width
,
meanline
=
True
,
showmeans
=
True
,
patch_artist
=
True
))
boxplots
.
append
(
ax
.
boxplot
(
recall
,
positions
=
pos
,
widths
=
width
,
meanline
=
True
,
showmeans
=
True
,
boxplots
.
append
(
ax
.
boxplot
(
recall
.
transpose
()
,
positions
=
pos
,
widths
=
width
,
meanline
=
True
,
showmeans
=
True
,
patch_artist
=
True
))
boxplots
.
append
(
ax
.
boxplot
(
accuracy
,
positions
=
pos
+
1.5
*
width
,
widths
=
width
,
meanline
=
True
,
boxplots
.
append
(
ax
.
boxplot
(
accuracy
.
transpose
()
,
positions
=
pos
+
1.5
*
width
,
widths
=
width
,
meanline
=
True
,
showmeans
=
True
,
patch_artist
=
True
))
boxplots
.
append
(
ax
.
boxplot
(
auroc
,
positions
=
pos
+
3
*
width
,
widths
=
width
,
meanline
=
True
,
boxplots
.
append
(
ax
.
boxplot
(
auroc
.
transpose
()
,
positions
=
pos
+
3
*
width
,
widths
=
width
,
meanline
=
True
,
showmeans
=
True
,
patch_artist
=
True
))
count
=
0
colors
=
[
'
red
'
,
'
yellow
'
,
'
blue
'
,
'
tan
'
,
'
green
'
]
...
...
This diff is collapsed.
Click to expand it.
Snakefile
+
20
−
39
View file @
b2e66543
configfile: 'config.yaml'
import ANN_Data_Generator, Initial_Condition, ANN_Training
from ANN_Training import evaluate_models
import numpy as np
def replace_none(list):
return {} if list is None else list
DIR = config['data_directory']
MODELS = config['models']
if config['random_seed'] is not None:
np.random.seed(config['random_seed'])
rule all:
input:
DIR+'/trained models/model__
' + config['model_name'] + '.pt'
,
DIR+'/model evaluation/classification_accuracy/' +
config['model_name']
+ '.pdf'
expand(
DIR+'/trained models/model__
{model}.pt', model=MODELS)
,
DIR+'/model evaluation/classification_accuracy/' +
'_'.join(MODELS.keys())
+ '.pdf'
rule test_model:
input:
DIR+'/input_data.npy',
DIR+'/output_data.npy'
params:
model_name = config['model_name'],
num_epochs = config['num_epochs'],
threshold = config['threshold'],
batch_size = config['batch_size'],
model = config['model'],
model_config = replace_none(config['model_config']),
loss_function = config['loss_function'],
optimizer = config['optimizer']
log:
DIR+'/log/test_model.log'
output:
DIR+'/model evaluation/classification_accuracy/' +
config['model_name']
+ '.pdf'
DIR+'/model evaluation/classification_accuracy/' +
'_'.join(MODELS.keys())
+ '.pdf'
run:
trainer= ANN_Training.ModelTrainer({'model_name': params.model_name,
'num_epochs': params.num_epochs, 'dir': DIR,
'model_dir': DIR, 'threshold': params.threshold,
'batch_size': params.batch_size, 'model': params.model,
'model_config': params.model_config,
'loss_function': params.loss_function,
'optimizer': params.optimizer})
trainer.test_model()
models = {}
for model in MODELS:
trainer= ANN_Training.ModelTrainer({'model_name': model, 'dir': DIR,
'model_dir': DIR, **MODELS[model]})
models[model] = trainer
evaluate_models(models, DIR, 2)
rule generate_data:
output:
...
...
@@ -74,26 +65,16 @@ rule train_model:
DIR+'/input_data.npy',
DIR+'/output_data.npy'
params:
model_name = config['model_name'],
num_epochs = config['num_epochs'],
threshold = config['threshold'],
batch_size = config['batch_size'],
model = config['model'],
model_config = replace_none(config['model_config']),
loss_function = config['loss_function'],
optimizer = config['optimizer']
models = MODELS
log:
DIR+'/log/train_model.log'
output:
DIR+'/trained models/model__
' + config['model_name'] + '.pt'
,
DIR+'/trained models/loss__
' + config['model_name'] + '.pt'
expand(
DIR+'/trained models/model__
{model}.pt', model=MODELS)
,
expand(
DIR+'/trained models/loss__
{model}.pt', model=MODELS)
run:
trainer= ANN_Training.ModelTrainer({'model_name': params.model_name,
'num_epochs': params.num_epochs, 'dir': DIR,
'model_dir': DIR, 'threshold': params.threshold,
'batch_size': params.batch_size, 'model': params.model,
'model_config': params.model_config,
'loss_function': params.loss_function,
'optimizer': params.optimizer})
for model in params.models:
print(model)
trainer= ANN_Training.ModelTrainer({'model_name': model, 'dir': DIR,
'model_dir': DIR, **params.models[model]})
trainer.epoch_training()
trainer.save_model()
\ No newline at end of file
This diff is collapsed.
Click to expand it.
config.yaml
+
17
−
8
View file @
b2e66543
...
...
@@ -24,12 +24,21 @@ functions:
adjustment
:
0
# Parameter for Model Training
model_name
:
Test_Name
models
:
Adam
:
num_epochs
:
1000
threshold
:
1.0e-5
batch_size
:
500
model
:
ThreeLayerReLu
model_config
:
model_config
:
{}
loss_function
:
BCELoss
optimizer
:
Adam
SGD
:
num_epochs
:
1000
threshold
:
1.0e-5
batch_size
:
500
model
:
ThreeLayerReLu
model_config
:
{}
loss_function
:
BCELoss
optimizer
:
SGD
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment