| Titre : | Building Automated Machine Learning Platform | | Type de document : | projet fin études | | Auteurs : | EZZAT Mahmoud, Auteur | | Langues : | Anglais (eng) | | Catégories : | Ingénierie de web et Informatique mobile
| | Mots-clés : | Keywords:
v Data science
v Machine Learning
v AutoML
v Preprocessing
v Data science pipeline
v Hyper parameter
v Model selection
v Tuning
v Meta-learning
v Feature engineering
v Feature extraction
v Feature Selection
v Outperform | | Index. décimale : | 1930/18 | | Résumé : | Data Science needs powerful and effective methods and tools that are comprehensive and easy
to use. Machine learning methods are particularly important for generating useful knowledge
and information from rapidly growing amounts of data. However, their use is not always easy.
Fully or partially automating a machine learning process is a long-standing goal that has
become increasingly important in the last two years. And the success of Machine Learning in a
wide range of applications has led to an ever-increasing demand for machine learning systems
that can be used on the shelf by non-experts. To be effective in practice, such systems must
automatically choose a good algorithm and include preprocessing steps for a new data set, and
also define their respective hyper parameters. Recent work has begun to address the automatic
problem of machine learning (AutoML) using efficient Bayesian optimization methods. Based
on this, Auto SK-Learn introduced a new scikit-learn-based AutoML system, (using more than
15 classifiers and more than 14 entity preprocessing methods, and more than 4 preprocessing
methods). Our system, automatically takes into account past performance on similar datasets,
and builds sets from models evaluated during optimization, and automates not only the
recommendation part of the model but the entire Data pipeline. Science.
And to achieve peak performance, modern machine learning techniques require careful data
preprocessing and hyper parametric adjustment. In addition, with the increasing number of
machine learning models being developed, model selection is becoming increasingly important.
Automating the selection and tuning of machine learning pipelines composed of data
preprocessing methods and machine learning models has long been one of the goals of the
machine learning community. In this talk, the idea is to tackle this meta-learning task by
exploiting experiments carried out in some of the different datasets to guide the exploration of
possible pipeline space. In our case, we operate information gathered from several related
experiences.
We build and automate the machine learning platform to build, validate, and deploy models,
with data preprocessing, feature engineering, future extraction, and future selection methods
that make the dataset usable for machine learning. the predictive performance of the final
machine learning model.
And since many of these steps and practices exceed the capabilities of non-experts, AutoML is
proposed as an artificial intelligence-based solution to the ever-increasing challenge of the
Graduation Project Thesis
6
Machine Learning application. Automating the end-to-end process of the machine learning
application has the advantage of producing simpler solutions, creating these solutions faster,
and producing models that often outperform hand-crafted models.
|
Building Automated Machine Learning Platform [projet fin études] / EZZAT Mahmoud, Auteur . - [s.d.]. Langues : Anglais ( eng) | Catégories : | Ingénierie de web et Informatique mobile
| | Mots-clés : | Keywords:
v Data science
v Machine Learning
v AutoML
v Preprocessing
v Data science pipeline
v Hyper parameter
v Model selection
v Tuning
v Meta-learning
v Feature engineering
v Feature extraction
v Feature Selection
v Outperform | | Index. décimale : | 1930/18 | | Résumé : | Data Science needs powerful and effective methods and tools that are comprehensive and easy
to use. Machine learning methods are particularly important for generating useful knowledge
and information from rapidly growing amounts of data. However, their use is not always easy.
Fully or partially automating a machine learning process is a long-standing goal that has
become increasingly important in the last two years. And the success of Machine Learning in a
wide range of applications has led to an ever-increasing demand for machine learning systems
that can be used on the shelf by non-experts. To be effective in practice, such systems must
automatically choose a good algorithm and include preprocessing steps for a new data set, and
also define their respective hyper parameters. Recent work has begun to address the automatic
problem of machine learning (AutoML) using efficient Bayesian optimization methods. Based
on this, Auto SK-Learn introduced a new scikit-learn-based AutoML system, (using more than
15 classifiers and more than 14 entity preprocessing methods, and more than 4 preprocessing
methods). Our system, automatically takes into account past performance on similar datasets,
and builds sets from models evaluated during optimization, and automates not only the
recommendation part of the model but the entire Data pipeline. Science.
And to achieve peak performance, modern machine learning techniques require careful data
preprocessing and hyper parametric adjustment. In addition, with the increasing number of
machine learning models being developed, model selection is becoming increasingly important.
Automating the selection and tuning of machine learning pipelines composed of data
preprocessing methods and machine learning models has long been one of the goals of the
machine learning community. In this talk, the idea is to tackle this meta-learning task by
exploiting experiments carried out in some of the different datasets to guide the exploration of
possible pipeline space. In our case, we operate information gathered from several related
experiences.
We build and automate the machine learning platform to build, validate, and deploy models,
with data preprocessing, feature engineering, future extraction, and future selection methods
that make the dataset usable for machine learning. the predictive performance of the final
machine learning model.
And since many of these steps and practices exceed the capabilities of non-experts, AutoML is
proposed as an artificial intelligence-based solution to the ever-increasing challenge of the
Graduation Project Thesis
6
Machine Learning application. Automating the end-to-end process of the machine learning
application has the advantage of producing simpler solutions, creating these solutions faster,
and producing models that often outperform hand-crafted models.
|
|
193/00 
Faire une suggestion Affiner la recherche
Building Automated Machine Learning Platform / EZZAT Mahmoud
