Preparing Predictive Analysis Data

Before using ModelArts to build a predictive analytics model, upload data to OBS. The OBS bucket and ModelArts must be in the same region. For example, if the OBS bucket is in the CN North-Beijing4 region, ensure that the ModelArts management console is also in the CN North-Beijing4 region. Otherwise, data cannot be obtained.

Requirements on Datasets

The data set used in the predictive analytics project must be a table dataset in .csv format. For details about the table dataset, see Creating a Dataset.

To convert the data from .xlsx to .csv, perform the following operations:

Save the original table data in .xlsx. Choose File > Save As, select a local address, set Save as type: to CSV (Comma delimited), and click Save. Then, click OK in the displayed dialog box.

Requirements on the training data:

The number of columns in the training data must be the same, and there has to be at least 100 data records (a feature with different values is considered as different data records).
The training columns cannot contain timestamp data (such as yy-mm-dd or yyyy-mm-dd).
If a column has only one value, the column is considered invalid. Ensure that there are at least two values in the label column and no data is missing.

The label column is the training target specified in a training task. It is the output (prediction item) for the model trained using the dataset.
In addition to the label column, the dataset must contain at least two valid feature columns. Ensure that there are at least two values in each feature column and that the percentage of missing data must be lower than 10%.
Due to the limitation of the feature filtering algorithm, place the predictive data column at the last. Otherwise, the training may fail.

Example of a table dataset:

The following table takes the bank deposit predictive dataset as an example. Data sources include age, occupation, marital status, cultural level, and whether there is a personal mortgage or personal loan.

**Table 1** Fields and meanings of data sources
Field	Meaning	Type	Description
attr_1	Age	Int	Age of the customer
attr_2	Occupation	String	Occupation of the customer
attr_3	Marital status	String	Marital status of the customer
attr_4	Education status	String	Education status of the customer
attr_5	Real estate	String	Housing situation of the customer
attr_6	Loan	String	Loan of the customer
attr_7	Deposit	String	Deposit of the customer

**Table 2** Sample data of the dataset
attr_1	attr_2	attr_3	attr_4	attr_5	attr_6	attr_7
31	blue-collar	married	secondary	yes	no	no
41	management	married	tertiary	yes	yes	no
38	technician	single	secondary	yes	no	no
39	technician	single	secondary	yes	no	yes
39	blue-collar	married	secondary	yes	no	no
39	services	single	unknown	yes	no	no

Uploading Data to OBS

In this section, the OBS console is used to upload data.

Upload files to OBS according to the following specifications:

The OBS path of the predictive analytics projects must comply with the following rules:

The OBS path of the input data must redirect to the data files. The data files must be stored in a folder in an OBS bucket rather than the root directory of the OBS bucket, for example, /obs-xxx/data/input.csv.
There must be at least 100 lines of valid data in .csv. There cannot be more than 200 columns of data and the total data size must be smaller than 100 MB.

Procedure for uploading data to OBS:

Perform the following operations to import data to the dataset for model training and building.

Log in to the OBS console and create a bucket in the same region as ModelArts. If an available bucket exists, ensure that the OBS bucket and ModelArts are in the same region.
Upload the local data to the OBS bucket. If you have a large amount of data, use OBS Browser+ to upload data or folders. The uploaded data must meet the dataset requirements of the ExeML project.

Upload data from unencrypted buckets. Otherwise, training will fail because data cannot be decrypted.

Creating a Dataset

After the data is prepared, create a proper dataset. For details, see Creating a Dataset.

FAQs

How do I process Schema information when creating a table dataset using data selected from OBS?

Schema information includes the names and types of table columns, which must be the same as those of the imported data.

If the original table contains a table header, enable Contain Table Header. The first row of the file will be used as column names. You do not need to modify the Schema information.
If the original table does not contain a table header, disable Contain Table Header. After data is selected from OBS, the column names will be used as the first row of the table by default. Change the column names to attr_1, attr_2, ..., attr_n. attr_n is the prediction column placed at last.