Difference between revisions of "Meteos/DatasetsandModels"

Revision as of 06:43, 4 December 2016

Meteos Dataset

Dataset is a data to create a prediction model.

Meteos currently supports following data format.

• CSV data format

<label>,<value1>,<value2>, ... <valueN>

• LibSVM data format

<label> <index1>:<value1> <index2>:<value2> ... <indexN>:<valueN>

When creating a prediction model, user specify a "source_dataset_url" parameter which show the place where dataset is located.

A Source_dataset_url has two url types as follows:

• Swift URL

Swift URL is used when creating a model from swift.

If it is no neccesary to parse a dataset, user can create a model from dataset in swift directly by specify source_data_url as below.

swift://<container_name>/<object_name>

• Internal HDFS URL

Internal HDFS URL is user when creating a modef from internal hdfs of meteos experiment. Dataset in internal hdfs has been already downloaded or parsed by Meteos.

When creating a model from dataset in hdfs, user have to specify a url as below.

internal://<dataset_id>

Meteos Prediction Model

Currently Meteos supports these following prediction models of Apache Spark.

Apache Spark has two machine learning libraries (MLlib and Ml).

MLlib and ML has multiple prediction models by data mining and machine learning algorithms.

MLlib

LinearRegression Model

LogisticRegression Model

DecisionTree Model

Kmeans Model

Recommendation Model

ML

Not Supported.