NTUA Seminar Talk: CATHERINE HUBER-CAROL - Comparing prediction ability of models and sub-models
Tue 8 May 2012 - 10:00
Ε Θ Ν Ι Κ Ο Μ Ε Τ Σ Ο Β Ι Ο Π Ο Λ Υ Τ Ε Χ Ν Ε Ι Ο
ΣΧΟΛΗ ΕΦΑΡΜΟΣΜΕΝΩΝ ΜΑΘΗΜΑΤΙΚΩΝ & ΦΥΣΙΚΩΝ ΕΠΙΣΤΗΜΩΝ
Τομέας Μαθηματικών
Πολυτεχνειούπολη – Zωγράφου ΑΘΗΝΑ - 157 80
ΤΗΛ. : 772 3291,
FAX : 772 1775
Δ Ι Α Λ Ε Ξ Η
Ομιλητής: CATHERINE HUBER-CAROL,
University Paris Descartes
Τίτλος : «Comparing prediction ability of models and sub-models»
Περίληψη: When the objective of modeling a data set is explanatory, it is most appropriate to choose the best fitting model using the usual model selection procedures. But if the objective is to predict and not to explain the facts, and some of the factors selected by the ‘ best fitting model’ are not available for all subjects, one can compare models by their prediction qualities rather than by their goodness of fit to the data. We consider here the case of two competing, nested, probability predicting models, consistent with Pencina et al. (2008) setting. The nested model contains traditional factors, and the larger model contains in addition some expensive, or generally hard to obtain, factor. We estimate the two nested models and two criteria comparing their predicting abilities on the same unique available sample and prove their asymptotic properties: consistency and normality.
Η ομιλία θα δοθεί την Παρασκευή 11 Μαΐου 2012 και ώρα 14:15, στην Αίθουσα Σεμιναρίων του Τομέα Μαθηματικών, κτ. Ε΄, 2ος όροφος.
Η Επιτροπή Σεμιναρίων
ΣΧΟΛΗ ΕΦΑΡΜΟΣΜΕΝΩΝ ΜΑΘΗΜΑΤΙΚΩΝ & ΦΥΣΙΚΩΝ ΕΠΙΣΤΗΜΩΝ
Τομέας Μαθηματικών
Πολυτεχνειούπολη – Zωγράφου ΑΘΗΝΑ - 157 80
ΤΗΛ. : 772 3291,
FAX : 772 1775
Δ Ι Α Λ Ε Ξ Η
Ομιλητής: CATHERINE HUBER-CAROL,
University Paris Descartes
Τίτλος : «Comparing prediction ability of models and sub-models»
Περίληψη: When the objective of modeling a data set is explanatory, it is most appropriate to choose the best fitting model using the usual model selection procedures. But if the objective is to predict and not to explain the facts, and some of the factors selected by the ‘ best fitting model’ are not available for all subjects, one can compare models by their prediction qualities rather than by their goodness of fit to the data. We consider here the case of two competing, nested, probability predicting models, consistent with Pencina et al. (2008) setting. The nested model contains traditional factors, and the larger model contains in addition some expensive, or generally hard to obtain, factor. We estimate the two nested models and two criteria comparing their predicting abilities on the same unique available sample and prove their asymptotic properties: consistency and normality.
Η ομιλία θα δοθεί την Παρασκευή 11 Μαΐου 2012 και ώρα 14:15, στην Αίθουσα Σεμιναρίων του Τομέα Μαθηματικών, κτ. Ε΄, 2ος όροφος.
Η Επιτροπή Σεμιναρίων
- UNIVERSITY OF PIRAEUS - SEMINAR 17/6/2015: Single and Multi Index Models for Mean and Quantile Regression
- NTUA Seminars: Statistical Inference for non-Markovian Stochastic Volatility Models by Alexandra Chronopoulou (Univ. of Illinois at Urbana-Champaign)
- Deep Learning AUEB Seminar - Wed 14 Oct 2015: Representation and deep learning with Bayesian non-parametric models
- NTUA Seminar: Ομάδες με ομολογική διάσταση 1 από τον Γιάννη Εμμανουήλ (ΕΚΠΑ)
- NTUA Seminar: "Towards a Syntactic/Probabilistic Framework for Vision/Speech Recognition and Biological Processes" by Vasilis Gidas
Permissions in this forum:
You cannot reply to topics in this forum
|
|