VictorMylle/Thesis
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Updated thesis

Browse Source
This commit is contained in:
Victor Mylle
2024-05-18 11:58:45 +02:00
parent 8a219d0d19
commit 5c365ebd88
8 changed files with 16 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Reports/Thesis/sections/results/policies/nrv_samples_policy.tex
View File

@@ -1,5 +1,7 @@
\subsection{Policy using generated NRV samples}
The generated full-day samples can be used to improve the profit of the policy. For each day, the generated samples can be used to determine the buying and selling thresholds. Assume that there are 100 generated full-day NRV samples for the day for which the profit needs to be optimized. The thresholds are determined for each generated sample separately using a simple grid search. All these thresholds can then be reduced by taking the mean to get one value for the buying threshold and one value for the selling threshold. Again, the penalty parameter is optimized for the test set to make sure around 283 charge cycles are used for fair comparison. The policy is evaluated for the different types of models that were trained and discussed in the previous sections. To reduce the computational cost, the policy is only evaluated for the best-performing models based on the CRPS metric.
The generated full-day samples can be used to improve the profit of the policy. For each day, the generated samples can be used to determine the buying and selling thresholds. Assume that there are 100 generated full-day NRV samples for the day for which the profit needs to be optimized. The thresholds are determined for each generated sample separately using a simple grid search. All these thresholds can then be reduced by taking the mean to get one value for the buying threshold and one value for the selling threshold. Again, the penalty parameter is optimized for the test set to make sure around 283 charge cycles are used for fair comparison.
A low CRPS value does not necessarily mean the policy will generate a high profit. Because of this, the CRPS metric can not be used to evaluate the model during the training phase and use this metric to do early stopping. To fairly evaluate and compare the models, a validation set is split off from the training set. The validation set is used to evaluate the profit of the policy during the training and use this to do early stopping. The last two months of the training set are used as the validation set. This range starts on 01-11-2022 and ends on 31-12-2022. Two months are chosen to make sure enough data is available to have a good estimate of the profit while making sure the validation set is not too large. The policy can be evaluated quite fast on the validation set which is feasible to do during the training after a certain number of epochs.
\begin{table}[ht]
\centering

4
Reports/Thesis/verslag.aux
View File

@@ -121,8 +121,8 @@
\@writefile{lot}{\contentsline {table}{\numberline {11}{\ignorespaces Results of the baseline policies on the test set. \relax }}{45}{table.caption.29}\protected@file@percent }
\newlabel{tab:fixed_thresholds}{{11}{45}{Results of the baseline policies on the test set. \relax }{table.caption.29}{}}
\@writefile{toc}{\contentsline {subsection}{\numberline {7.2}Policy using generated NRV samples}{45}{subsection.7.2}\protected@file@percent }
\@writefile{lot}{\contentsline {table}{\numberline {12}{\ignorespaces Comparison of the different models using the profit metric. The best-performing models for a certain type are selected based on the profit.\relax }}{45}{table.caption.30}\protected@file@percent }
\newlabel{tab:policy_comparison}{{12}{45}{Comparison of the different models using the profit metric. The best-performing models for a certain type are selected based on the profit.\relax }{table.caption.30}{}}
\@writefile{lot}{\contentsline {table}{\numberline {12}{\ignorespaces Comparison of the different models using the profit metric. The best-performing models for a certain type are selected based on the profit.\relax }}{46}{table.caption.30}\protected@file@percent }
\newlabel{tab:policy_comparison}{{12}{46}{Comparison of the different models using the profit metric. The best-performing models for a certain type are selected based on the profit.\relax }{table.caption.30}{}}
\abx@aux@page{6}{48}
\abx@aux@page{7}{48}
\abx@aux@page{8}{48}

4
Reports/Thesis/verslag.log
View File

@@ -1,4 +1,4 @@
This is pdfTeX, Version 3.141592653-2.6-1.40.25 (TeX Live 2023) (preloaded format=pdflatex 2023.9.17) 15 MAY 2024 16:18
This is pdfTeX, Version 3.141592653-2.6-1.40.25 (TeX Live 2023) (preloaded format=pdflatex 2023.9.17) 18 MAY 2024 11:50
entering extended mode
restricted \write18 enabled.
file:line:error style messages enabled.
@@ -2007,7 +2007,7 @@ Here is how much of TeX's memory you used:
1141 hyphenation exceptions out of 8191
84i,16n,131p,2100b,5180s stack positions out of 10000i,1000n,20000p,200000b,200000s
</Users/victormylle/Library/texlive/2023/texmf-var/fonts/pk/ljfour/public/bbm/bbm12.600pk></usr/local/texlive/2023/texmf-dist/fonts/type1/public/libertinust1math/LibertinusT1Math.pfb></usr/local/texlive/2023/texmf-dist/fonts/type1/public/libertine/LinBiolinumT.pfb></usr/local/texlive/2023/texmf-dist/fonts/type1/public/libertine/LinBiolinumTB.pfb></usr/local/texlive/2023/texmf-dist/fonts/type1/public/libertine/LinLibertineT.pfb></usr/local/texlive/2023/texmf-dist/fonts/type1/public/libertine/LinLibertineTB.pfb></usr/local/texlive/2023/texmf-dist/fonts/type1/public/libertine/LinLibertineTI.pfb></usr/local/texlive/2023/texmf-dist/fonts/type1/public/stix/stix-mathcal.pfb>
Output written on verslag.pdf (51 pages, 8683971 bytes).
Output written on verslag.pdf (51 pages, 8684536 bytes).
PDF statistics:
672 PDF objects out of 1000 (max. 8388607)
497 compressed objects within 5 object streams

BIN
Reports/Thesis/verslag.pdf
View File

Binary file not shown.

0
Reports/Thesis/verslag.synctex(busy)
View File

BIN
Reports/Thesis/verslag.synctex.gz
View File

Binary file not shown.

10
src/trainers/trainer.py
View File

@@ -114,8 +114,8 @@ class Trainer:
predict_sequence_length=self.model.output_size, full_day_skip=True
)
train_loader, test_loader = self.data_processor.get_dataloaders(
predict_sequence_length=self.model.output_size
train_loader, val_loader, test_loader = self.data_processor.get_dataloaders(
predict_sequence_length=self.model.output_size, validation=True
)
train_samples = self.random_samples(train=True, num_samples=5)
@@ -146,7 +146,7 @@ class Trainer:
running_loss += loss.item()
running_loss /= len(train_loader.dataset)
test_loss = self.test(test_loader)
test_loss = self.test(val_loader)
if self.patience is not None:
if (
@@ -170,7 +170,7 @@ class Trainer:
)
task.get_logger().report_scalar(
title=self.criterion.__class__.__name__,
series="test",
series="val",
value=test_loss,
iteration=epoch,
)
@@ -194,7 +194,7 @@ class Trainer:
# )
if hasattr(self, "calculate_crps_from_samples"):
self.calculate_crps_from_samples(task, test_loader, epoch)
self.calculate_crps_from_samples(task, val_loader, epoch)
if task:
self.finish_training(task=task)

8
src/training_scripts/diffusion_training.py
View File

@@ -2,9 +2,9 @@ from src.utils.clearml import ClearMLHelper
clearml_helper = ClearMLHelper(project_name="Thesis/NrvForecast")
task = clearml_helper.get_task(
task_name="Diffusion Training: hidden_sizes=[512, 512] (100 steps), lr=0.0001, time_dim=8",
task_name="Diffusion Training: hidden_sizes=[1024, 1024] (300 steps), all features",
)
# task.execute_remotely(queue_name="default", exit_process=True)
task.execute_remotely(queue_name="default", exit_process=True)
from src.models import *
from src.losses import *
@@ -42,7 +42,7 @@ print("Input dim: ", inputDim)
model_parameters = {
"epochs": 15000,
"learning_rate": 0.0001,
"hidden_sizes": [512, 512],
"hidden_sizes": [1024, 1024],
"time_dim": 8,
}
@@ -71,6 +71,6 @@ policy_evaluator = PolicyEvaluator(baseline_policy, task)
#### Trainer ####
trainer = DiffusionTrainer(
model, data_processor, "cuda", policy_evaluator=policy_evaluator, noise_steps=20
model, data_processor, "cuda", policy_evaluator=policy_evaluator, noise_steps=300
)
trainer.train(model_parameters["epochs"], model_parameters["learning_rate"], task)