Dataset Clause Examples

Dataset. Information or data from a database Service primarily devoted to market sizing by market segment and country, derived from a single (a) market segment and (b) country.

Dataset. 4.1. The Dataset will consist of all the files transferred by the Depositor and the metadata provided by the Depositor. Metadata is understood to mean the contents of all fields that must be completed in the archival system at the time of deposit in order to describe the Dataset. 4.2. The Depositor declares that the Dataset corresponds to the metadata provided by the Depositor. 4.3. The Dataset shall be compiled with due observance of the Netherlands Code of Conduct for Research Integrity, the GDPR and other applicable laws and regulations. 4.4. The metadata and file names shall not contain any personal data within the meaning of the GDPR. Only Bibliographical Data are allowed. It is explicitly forbidden to include personal data that are part of the deposited data, such as - but not limited to - research subjects, in the metadata and file names. 4.5. The Depositor will provide the files in a preferred format, as defined on the Depositary's website at the time of deposit. In the event that a format is not defined as a preferred format, the Depositor will contact the Depositary before delivery. A different file format may only be supplied with the written consent of the Depositary. 4.6. The Depositor will provide documentation with the Dataset that explains its creation, contents and any specific values (such as codes, characters and abbreviations), its structure (such as folder structures and relationships between files) and its actual use (such as that of software) to third parties. 4.7. The Depositor will make the Dataset available to the Depositary in a manner and through a medium that the Depositary deems suitable.

Dataset. The database we used has been developed within the subalpine GIG. Samples were collected between 1997 and 2010 in the sublittoral zone of 19 Austrian, 25 German, 21 France and 28 Italian subalpine lakes. 10 of those lakes were sampled in 2 different years while 1 lake was sampled in 3 different years, each lake-year combination has been considered as an indipendent sample unit. Invertebrates were indentifyed to the lower taxonomic level possible, mostly to genus/species level. Data gathered in more than one sampling site were aggregated to lake-year level. We considered climatic and morphological environmental variables (table 1): precipitation, mean annual temperature, difference between temperature in July and in January, lake surface area, lake mean depth and catchment area. The climatic data were gathered from the Climatic Research Unit (CRU) model (New et al. 2002; ▇▇▇▇://▇▇▇.▇▇▇.▇▇▇.▇▇.▇▇/). Table 1: Environmental variables ranges. min max Mean annual Prec. (cm) 60.16 162.67 Mean annual Temp. (°C) 5.17 12.99 T(July)-T(January) (°C) 17.40 21.60 surface (km2) 0.04 79.90 mean depth (m) 3.20 53.21 catchment (km2) 1.01 4551.60

Dataset. Any data you provide to the Project is subject to the license agreement indicated in the Project’s source repository for the Materials.

Dataset. Dataset description

Dataset. Figure 1: Excerpt from the AffectNet (▇▇▇▇▇▇▇▇▇▇▇▇ 2015) database. High quality emotional annotations (i.e. continuous valence-arousal dimensions) are more commonly available for facial expressions than for multi-modal data. In contrast to databases containing multi-modal emotional expressions (i.e. including high quality spoken audio), well annotated data exclusively featuring facial expressions are more broadly available. The field of image processing is by far the most advanced research topic concerning deep learning approaches. Classification problems are accessible due to their visual nature and the number of adequately annotated databases for a wide range of recognition tasks is relatively high. However, the majority of huge databases are focused on object recognition. Though there is a selection of publicly available databases, affect recognition (primarily from facial expressions) is not the most common image processing topic. Our need for high quality annotations in the valence - arousal space further limits the selection, as databases featuring only categorical emotion labels are more common. Therefore our current model is again based on transfer learning techniques and is for now mainly trained on a single comprehensive affective corpus called AffectNet (▇▇▇▇▇▇▇▇▇▇▇▇, 2015). The collection of facial expressions consists of close-up shots of human faces taken from publicly available video and picture collections (reference). Annotations contain categorical emotion labels as well as continuous valence - arousal scores.

Dataset. For Chinese-English (ZH-EN) translation, our training data for the translation task consists of 1.25M Chinese-English sentence pairs extracted from LDC corpora1. The NIST02 testset is chosen as the development set, and the ▇▇▇▇▇▇, ▇▇▇▇ corpora include LDC2002E18, LDC2003E07,LDC2003E14, Hansards portion of LDC2004T07, LDC2004T08 and LDC2005T06. # Model NIST WMT 1 EDR (Tu et al., 2017) N/A N/A 33.73 34.15 N/A N/A 2 DB (▇▇▇▇▇ et al., 2018) 38.02 40.83 ▇/▇ ▇/▇ ▇/▇ ▇/▇ 3 Transformer(Base) 45.57 46.40 46.11 44.92 45.75 27.28 4 +lossmse 46.71† 47.23† 47.12† 45.78† 46.71 28.11† 5 +lossmse + enhanced 46.94† 47.52† 47.43† 46.04† 46.98 28.38† 6 Transformer(Big) 46.73 47.36 47.15 46.82 47.01 28.36 7 +lossmse 47.43† 47.96 47.78 47.39 47.74 28.71 8 +lossmse + enhanced 47.68† 48.13† 47.96† 47.56† 47.83 28.92†

Dataset. The dataset is divided into two groups: native prose and translated prose. Native prose comprises the so-called Mabinogion corpus, found in the White Book of Rhydderch (NLW Peniarth 4 & 5, dated c. 1350) and the slightly later Red Book of Hergest (▇▇▇▇▇ College Oxford 111, dated c. 1385). This corpus consists of eleven narrative tales, usually dated ‘between the end of the eleventh and the beginning of the fourteenth centuries’; the details of their date remain debated (▇▇▇▇▇▇ 1998: 134; Rodway 2013: 1). The first four tales are known as the Pedeir Keinc, the ‘Four Branches’ (▇▇▇▇▇▇▇▇ 1930). These include narratives named after the four major characters: ▇▇▇▇▇, ▇▇▇▇▇▇▇, ▇▇▇▇▇▇▇▇▇ and Math. Then there are three Arthurian tales about ▇▇▇▇▇▇▇, ▇▇▇▇▇ and ▇▇▇▇▇▇▇, traditionally labelled the ‘Three Romances’ although this label is contested (see esp. ▇▇▇▇▇-▇▇▇▇▇▇ 2004). Arthurian literature of this kind featuring the same protagonists is found in other European languages, including ▇▇▇▇▇▇▇▇ ▇▇ ▇▇▇▇▇▇’s French versions. The relationship of the Welsh Romances to their French counterparts has been long debated, and the current consensus treats them as native compositions but ones influenced to some extent by the French ones (see ▇.▇. ▇▇▇▇▇-▇▇▇▇▇▇ 1991; see also ▇▇▇▇▇ & ▇▇▇▇ 2006 & 2008, ▇▇▇▇ 2010, and ▇▇▇▇▇-▇▇▇▇▇ 2014 on the Welsh Charlemagne cycle). We include the Romances tentatively in the native corpus, but will also discuss them separately in section 4 to see whether they differ significantly from the other native texts. We then have four further native tales: Culhwch ac Olwen, Breudwyt Maxen (‘The Dream of Macsen’), Breudwyt Ronabwy (‘The Dream of Rhonabwy’) and Cyfranc Lludd a Llefelys (‘The Tale/Encounter of Lludd and Llefelys’, which also occurs inserted into the NLW Llanstephan 1 version of Brut y Brenhinedd and other later versions).3 Culhwch ac Olwen is usually taken to be somewhat earlier linguistically than the other tales of the Mabinogion corpus (see Rodway 2013: 1, fn. 2) and we analyse this tale separately in section 4 to see whether it differs from the other texts as regards adjectival agreement. For our annotated corpus, only the edited versions of the White Book of Rhydderch texts have been used.4 1). The Bible translation dates from around the start of the Early Modern Welsh period, usually taken to begin at c. 1500 (▇▇▇▇▇▇▇ 1995: xviii). This, and also the mid-fourteenth- century Buchedd ▇▇▇▇, are therefore somewhat later in date than the nat...

Dataset. To generate the vast amount of training data needed to train a deep neural network from scratch, we employ a sophisticated method to automatically generate annotations: We looked for publicly available videos from television media libraries that feature high quality subtitles with timestamps (reference). With these timestamps we are able to automatically segment annotations for voiced audio parts against silence or various kinds of sounds we want to identify as background noises.