Yd
Y.C. de Vries
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
2 records found
1
Master thesis
(2021)
-
Y.C. de Vries, N. Yorke-Smith, E. Demirović, Christian van Ommeren, Jan-Pieter Paardekooper
With the increasing global demand for logistics, supply chains have grown a lot in volume over the last decades. To be able to operate effectively within the capacity constraints of the carriers, proper collaboration and optimization of order allocation is required. Van Berkel Logistics facilitates the transport of containers by trucks from sea terminals in Rotterdam to inland customers and back. This logistical planning problem is manually solved by planners on a daily basis. Within this research it is investigated to what extent reinforcement learning could be applied for solving this planning problem of moving containers in an automated way. A simulation environment was constructed which represents the container planning dynamics. It was made as accurate as reasonably possible with the help of historic data. Three reinforcement learning models, the OnePass, Iterative and Attention model, have been developed and tested for their ability to learn to choose proper orders such that the orders are as much on time as possible. A main challenge in constructing these models was to design them such that they could cope with a varying state and action space. In an experimental evaluation, it was found that the models are able to learn to make better decisions over time and eventually perform similar to the heuristic baseline tested out in terms of total lateness observed. In terms of driven distance and fraction on time orders, the OnePass and Iterative model were able to beat the heuristic choices. Overall, the Iterative model has shown the best performance and is able to learn scenarios as big as real-life scenarios van Berkel Logistics deals with. However, it also tends to be slower than the other models due to its iterative approach.
...
With the increasing global demand for logistics, supply chains have grown a lot in volume over the last decades. To be able to operate effectively within the capacity constraints of the carriers, proper collaboration and optimization of order allocation is required. Van Berkel Logistics facilitates the transport of containers by trucks from sea terminals in Rotterdam to inland customers and back. This logistical planning problem is manually solved by planners on a daily basis. Within this research it is investigated to what extent reinforcement learning could be applied for solving this planning problem of moving containers in an automated way. A simulation environment was constructed which represents the container planning dynamics. It was made as accurate as reasonably possible with the help of historic data. Three reinforcement learning models, the OnePass, Iterative and Attention model, have been developed and tested for their ability to learn to choose proper orders such that the orders are as much on time as possible. A main challenge in constructing these models was to design them such that they could cope with a varying state and action space. In an experimental evaluation, it was found that the models are able to learn to make better decisions over time and eventually perform similar to the heuristic baseline tested out in terms of total lateness observed. In terms of driven distance and fraction on time orders, the OnePass and Iterative model were able to beat the heuristic choices. Overall, the Iterative model has shown the best performance and is able to learn scenarios as big as real-life scenarios van Berkel Logistics deals with. However, it also tends to be slower than the other models due to its iterative approach.
Accurate diagnosis and treatment of patients infected with multiple strains of a pathogen is a challenging task. The use of whole genome sequencing techniques provide high potential to give proper insight into the microbial composition of human metagenomic samples. Distinguishing multiple strains of a certain species is difficult due to the high similarity in genetic content. Currently several tools aimed at the identification of different strains in metagenomic sequence data are available. We present an independent benchmark to compare the performance of several of these tools. The tools have been evaluated with a variety of synthetic metagenomic samples containing strain mixtures of the species Enterococcus, Escherichia coli and Mycobacterium tuberculosis.
To facilitate this research, a benchmark framework in Python 3 was built. This framework made it possible to test the performance of tools aiming at unraveling the composition of sequence data. It is able to automatically generate batches of metagenomic readsets with custom predefined properties. The tools can easily do their analysis on those reads in a streamlined fashion. The output of the tools are put in a standardized format to make the complete comparison of tools easier.
This framework has been built as part of our Bachelor End Project over the course of 10 weeks. In the first few weeks we became familiar with the domain of bioinformatics and the type of tools that had to be included in this research. The implementation of the framework required thorough understanding of the tools and took quite some time to implement. Towards the end of the project, the framework has been used to run the tools with a large variety of synthetic readsets. Analysis of these outputs resulted in an insightful overview of the tools capabilities as presented in this paper. ...
To facilitate this research, a benchmark framework in Python 3 was built. This framework made it possible to test the performance of tools aiming at unraveling the composition of sequence data. It is able to automatically generate batches of metagenomic readsets with custom predefined properties. The tools can easily do their analysis on those reads in a streamlined fashion. The output of the tools are put in a standardized format to make the complete comparison of tools easier.
This framework has been built as part of our Bachelor End Project over the course of 10 weeks. In the first few weeks we became familiar with the domain of bioinformatics and the type of tools that had to be included in this research. The implementation of the framework required thorough understanding of the tools and took quite some time to implement. Towards the end of the project, the framework has been used to run the tools with a large variety of synthetic readsets. Analysis of these outputs resulted in an insightful overview of the tools capabilities as presented in this paper. ...
Accurate diagnosis and treatment of patients infected with multiple strains of a pathogen is a challenging task. The use of whole genome sequencing techniques provide high potential to give proper insight into the microbial composition of human metagenomic samples. Distinguishing multiple strains of a certain species is difficult due to the high similarity in genetic content. Currently several tools aimed at the identification of different strains in metagenomic sequence data are available. We present an independent benchmark to compare the performance of several of these tools. The tools have been evaluated with a variety of synthetic metagenomic samples containing strain mixtures of the species Enterococcus, Escherichia coli and Mycobacterium tuberculosis.
To facilitate this research, a benchmark framework in Python 3 was built. This framework made it possible to test the performance of tools aiming at unraveling the composition of sequence data. It is able to automatically generate batches of metagenomic readsets with custom predefined properties. The tools can easily do their analysis on those reads in a streamlined fashion. The output of the tools are put in a standardized format to make the complete comparison of tools easier.
This framework has been built as part of our Bachelor End Project over the course of 10 weeks. In the first few weeks we became familiar with the domain of bioinformatics and the type of tools that had to be included in this research. The implementation of the framework required thorough understanding of the tools and took quite some time to implement. Towards the end of the project, the framework has been used to run the tools with a large variety of synthetic readsets. Analysis of these outputs resulted in an insightful overview of the tools capabilities as presented in this paper.
To facilitate this research, a benchmark framework in Python 3 was built. This framework made it possible to test the performance of tools aiming at unraveling the composition of sequence data. It is able to automatically generate batches of metagenomic readsets with custom predefined properties. The tools can easily do their analysis on those reads in a streamlined fashion. The output of the tools are put in a standardized format to make the complete comparison of tools easier.
This framework has been built as part of our Bachelor End Project over the course of 10 weeks. In the first few weeks we became familiar with the domain of bioinformatics and the type of tools that had to be included in this research. The implementation of the framework required thorough understanding of the tools and took quite some time to implement. Towards the end of the project, the framework has been used to run the tools with a large variety of synthetic readsets. Analysis of these outputs resulted in an insightful overview of the tools capabilities as presented in this paper.