136879 research outputs found
Sort by
Forecasting Ride-Hailing Demand in Urban Areas: A Deep Ensemble and Time Series Clustering Approach
No full textThis paper investigates the increasingly important task of forecasting demand for ride-hailing services, which have significantly disrupted traditional transportation models. Notably, the study concentrates on New York City's Yellow Cab service, which despite the surge in popularity of app-based services, continues to serve a substantial number of commuters. The study highlights the necessity of accurate demand prediction for efficient resource allocation, reduced wait times and improved user satisfaction. Traditional forecasting methods like Historical Average, Exponential Weighted Moving Averages, ARIMA etc., are examined, alongside the more recent machine learning and data mining techniques, CNN-LSTM and XGBoost. A novel approach, utilizing an ensemble of machine learning models ��� XGBoost and Convolutional Neural Network ��� LSTM along with creative feature engineering is proposed for real-time demand forecasting across numerous locations. Furthermore, the study also tries to understand the application of time-series clustering methods and their effectiveness in grouping similar time-series together and extracting clustering features to improve the performance of the model. Additionally, the study observes the ineffectiveness of generalized model to forecast demand in low-demand reasons and presents possible research direction for solving the issue. This study contributes to the growing literature on demand forecasting in the ride-hailing industry and provides insights into the use of time-series clustering for the same
Synthesizable and Optimized Molecular Design With Variational Autoencoder
No full textThe use of generative models such as variational autoencoders (VAEs) for molecular design and optimization has gained popularity in recent years due to its efficiency in exploring high-dimensional molecular space to identify molecules with desired properties. However, these models often face challenges in producing molecules that are synthesizable. On the other hand, there exist synthesizable decoding models that can decode the synthetic route of a molecule using known reactants and reaction templates and translate it into its synthesizable self. But these models are not suitable to generate optimized and diverse molecules. In this work, we aim to combine the strengths of the VAE and synthesizable decoding models to overcome the shortcomings of utilizing the models separately. We proposed a novel approach that can address the limitations of existing molecule generation models, specifically focusing on generating diverse and property optimized molecules while maintaining synthesizability. The proposed method adopts training of a baseline model with a molecular dataset that satisfies some predefined constraints to produce property optimized molecules. The synthetic accessible (SA) score and the average dissimilarity among the generated molecules shows that the proposed method is also successful in producing synthesizable and diverse molecules
The Impacts of Furrow Diking, Terracing and Contour Cultivation on Water Conservation in Texas Agriculture
No full textWhere Does That Masked Man Live?
No full text"Published by, researched by, compiled by...Janet Ellicott."Not dated."An undated 24-page comprehensive listing of fanclubs, fanzines (mostly in the UK), contact and agents' addresses for a great many stars, tv shows and films." (Fanlore.org wiki