Your First Safe-DS Regression Program¶

Usually it's suggested to start your learning journey by printing "Hello World!", but since Safe-DS is a language for data science, we are going to step further with this short tutorial: You will learn how to write your first Safe-DS pipeline, which will import data from a CSV file, train a regression model and print the mean squared error of that fitted model.

File¶

Start by creating a file with the extension .sds.

Package¶

All Safe-DS programs must declare their packages at the beginning of the file.

package regression

Pipeline¶

Next you have to define your pipeline which is the main part of your program. Pipelines are data science programs designed to solve a specific task. They act as the entry point to start execution.

pipeline demo {
  // All further code must go here
}

Reading Data¶

Using data from other files can be done easily by placing your data files in the same folder as your .sds file. It's common to use .csv files for data, so we will use them as well. Once you placed your file into that folder, you can write the following line to create a table.

val rawData = Table.fromCsvFile("input.csv");

Now you can use your data by accessing the variable rawData.

Cleaning Your Data¶

At this point is usual to clean your data. If your data needs no cleaning, you can skip this part.

Removing columns:

val removedColumns = rawData.removeColumns(["columnName"]);

You can now access your data by its new name removedColumns.

Removing rows:

val removedRows = removedColumns.removeRowsWithMissingValues();

You can also use an imputer in order to replace missing values with these different strategies:

SimpleImputer.Strategy.Mean (replace missing values with the mean of the column)
SimpleImputer.Strategy.Median (replace missing values with the median of the column)
SimpleImputer.Strategy.Mode (replace missing values with the most frequent value of the column)
SimpleImputer.Strategy.Constant (replace missing values with a constant value)

val columnsToImpute = ["columnName1", "columnName2"];
val fittedImputer = SimpleImputer(SimpleImputer.Strategy.Mean, columnsToImpute).fit(removedRows);
val imputedData = fittedImputer.transform(removedRows);

You can also use an label encoder to label specific values if your data is ordinal:

val columnsToEncode = ["columnName3", "columnName4"];
val fittedLabelEncoder = LabelEncoder(columnsToEncode).fit(imputedData);
val labeledData = fittedLabelEncoder.transform(imputedData);

It's also possible to use a one-hot encoder if your data is nominal:

val columnsToOneHotEncode = ["columnName5", "columnName6"];
val fittedOneHotEncoder = OneHotEncoder(columnsToOneHotEncode).fit(labeledData);
val oneHotEncodedData = fittedOneHotEncoder.transform(labeledData);

Create Training and Testing Set¶

In order to train a model on your data and later evaluate its performance, you have to create a test and training set:

val train, val test = oneHotEncodedData.splitRows(0.7);
val trainSet = train.toTabularDataset("targetColumnName");
val testSet = test.toTabularDataset("targetColumnName");

Regressor¶

Now you can train the data using a regressor: Here you can find other regressors. Create your regressor simply by initializing and assigning it to a variable.

val regressor = DecisionTreeRegressor();

Fitting a Regressor¶

Once you have defined your regressor, you can fit that regressor by calling the fit function and passing your train set.

val fittedRegressor = regressor.fit(trainSet);

Predicting with the Fitted Regressor¶

Now you can use your regressor to predict target values on your test set:

val _prediction = fittedRegressor.predict(testSet);

Evaluating the Fitted Regressor¶

Last but not least you might want to test evaluate model, which is easily achieved by these lines:

val _metrics = fittedRegressor.summarizeMetrics(testSet);

Full Code¶

package regression

pipeline demo {
    // Read data
    val rawData = Table.fromCsvFile("input.csv");

    // Remove columns and rows
    val removedColumns = rawData.removeColumns(["columnName"]);
    val removedRows = removedColumns.removeRowsWithMissingValues();

    // Impute
    val columnsToImpute = ["columnName1", "columnName2"];
    val fittedImputer = SimpleImputer(SimpleImputer.Strategy.Mean, columnsToImpute).fit(removedRows);
    val imputedData = fittedImputer.transform(removedRows);

    // Label encoding
    val columnsToEncode = ["columnName3", "columnName4"];
    val fittedLabelEncoder = LabelEncoder(columnsToEncode).fit(imputedData);
    val labeledData = fittedLabelEncoder.transform(imputedData);

    // One-hot encoding
    val columnsToOneHotEncode = ["columnName5", "columnName6"];
    val fittedOneHotEncoder = OneHotEncoder(columnsToOneHotEncode).fit(labeledData);
    val oneHotEncodedData = fittedOneHotEncoder.transform(labeledData);

    // Split and tag data
    val train, val test = oneHotEncodedData.splitRows(0.7);
    val trainSet = train.toTabularDataset("targetColumnName");
    val testSet = test.toTabularDataset("targetColumnName");

    // Define and train a regressor
    val regressor = DecisionTreeRegressor();
    val fittedRegressor = regressor.fit(trainSet);

    // Predict and evaluate
    val _prediction = fittedRegressor.predict(testSet);
    val _metrics = fittedRegressor.summarizeMetrics(testSet);
}