Random Forests variable importance measures are often used to rank variables by their relevance to a classification problem and subsequently reduce the number of model inputs in high-dimensional data sets, thus increasing computational efficiency. However, as a result of the way that training data and predictor variables are randomly selected for use in constructing each tree and splitting each node, it is also well known that if too few trees are generated, variable importance rankings tend to differ between model runs. In this letter, we characterize the effect of the number of trees (ntree) and class separability on the stability of variable importance rankings and develop a systematic approach to define the number of model runs and/or trees required to achieve stability in variable importance measures. Results demonstrate that both a large ntree for a single model run, or averaged values across multiple model runs with fewer trees, are sufficient for achieving stable mean importance values. While the latter is far more computationally efficient, both the methods tend to lead to the same ranking of variables. Moreover, the optimal number of model runs differs depending on the separability of classes. Recommendations are made to users regarding how to determine the number of model runs and/or trees that are required to achieve stable variable importance rankings.
The rise of game development and game studies on university campuses prompts academic libraries to consider how to support teaching and research in this area. This article examines current issues and challenges in the development of game collections at academic libraries. The gaming ecosystem has become more complex and libraries may need to move beyond collections largely based on console video games. This article will advance the discussion by considering emerging issues to support access to the full range of games. The article will use examples from Carleton University Library, Ottawa, which has been developing a game collection since 2008.
The design and analysis of community-scale energy systems and incentives is a non-trivial task. The challenge of such undertakings is the well-documented uncertainty of building occupant behaviours. This is especially true in the residential sector, where occupants are given more freedom of activity compared to work environments. Further complicating matters is the dearth of available measured data. Building performance simulation tools are one approach to community energy analysis, however such tools often lack realistic models for occupant-driven demands, such as appliance and lighting (AL) loads. For community-scale analysis, such AL models must also be able to capture the temporal and inter-dwelling variation to achieve realistic estimates of aggregate electrical demand. This work adapts the existing Centre for Renewable Energy Systems Technology (CREST) residential energy model to simulate Canadian residential AL demands. The focus of the analysis is to determine if the daily, seasonal, and inter-dwelling variation of AL demands estimated by the CREST model is realistic. An in-sample validation is conducted on the model using 22 high-resolution measured AL demand profiles from dwellings located in Ottawa, Canada. The adapted CREST model is shown to broadly capture the variation of AL demand variations observed in the measured data, however seasonal variation in daily AL demand behaviour was found to be under-estimated by the model. The average and variance of daily load factors was found to be similar between measured and modelled. The model was found to under-predict the daily coincidence factors of aggregated demands, although the variance of coincident factors was shown to be similar between measured and modelled. A stochastic baseload input developed for this work was found to improve estimates of the magnitude and variation of both baseload and peak demands.
This article describes the progress made toward implementing Resource Description and Access (RDA) in libraries across Canada, as of Fall 2013. Differences in the training experiences in the English-speaking cataloging communities and French-speaking cataloging communities are discussed. Preliminary results of a survey of implementation in English-Canadian libraries are included as well as a summary of the support provided for French-Canadian libraries. Data analysis includes an examination of the rate of adoption in Canada by region and by sector. Challenges in RDA training delivery in a Canadian context are identified, as well as opportunities for improvement and expansion of RDA training in the future.
It has been observed in the literature that as the cardinality of the prescribed discrete input-output data set increases, the corresponding four-bar linkages that minimise the Euclidean norm of the design and structural errors tend to converge to the same linkage. The important implication is that minimising the Euclidean norm, or any p-norm, of the structural error, which leads to a nonlinear least-squares problem requiring iterative solutions, can be accomplished implicitly by minimising that of the design error, which leads to a linear least-squares problem that can be solved directly. Apropos, the goal of this paper is to take the first step towards proving that as the cardinality of the data set tends towards infinity the observation is indeed true. In this paper we will integrate the synthesis equations in the range between minimum and maximum input values, thereby reposing the discrete approximate synthesis problem as a continuous one. Moreover, we will prove that a lower bound of the Euclidean norm, and indeed of any p-norm, of the design error for planar RRRR function-generating linkages exists and is attained with continuous approximate synthesis.
Building Performance Simulation (BPS) is a powerful tool to estimate and reduce building energy consumption at the design stage. However, the true potential of BPS remains unrealized if trial and error simulation methods are practiced to identify combinations of parameters to reduce energy use of design alternatives. Optimization algorithms coupled with BPS is a process-orientated tool which identifies optimal building configurations using conflicting performance indicators. However, the application of optimization approaches to building design is not common practice due to time and computation requirements. This paper proposes a hybrid evolutionary algorithm which uses information gained during previous simulations to expedite and improve algorithm convergence using targeted deterministic searches. This technique is applied to a net-zero energy home case study to optimize trade-offs in passive solar gains and active solar generation using a cost constraint.
Oral narrative skills are assumed to develop through parent-child interactive routines. One such
routine is shared reading. A causal link between shared reading and narrative knowledge,
however, has not been clearly established. The present research tested whether an 8-week
shared-reading intervention enhanced the fictional narrative skills of children entering formal
education. Dialogic reading, a shared reading activity that involves elaborative questioning
techniques, was used to engage children in oral interaction during reading and to emphasize
elements of story knowledge. Forty English-speaking five- and six-year-olds were assigned to
either the dialogic-reading or an alternative-treatment group. ANCOVA results found that the
dialogic-reading children’s post-test narratives were significantly better on structure and context
measures than those for the alternative-treatment children, but results differed for produced or
retold narratives. The dialogic-reading children also showed expressive vocabulary gains.
Overall, this study concretely determined that aspects of fictional narrative construction
knowledge can be learned from interactive book reading.