sklearn Component Library

The expops-sklearn package provides pre-built pipeline steps for common scikit-learn workflows: loading CSV data, splitting datasets, and training/evaluating estimators.

Installation

Install into the same environment as the expops CLI:

pip install expops-sklearn

Available Components

sklearn.csv_to_xy

Loads a CSV file and splits it into feature matrix X and label vector y.

Outputs: X, y

- name: "data_source"
  component: "sklearn.csv_to_xy"
  parameters:
    label_column: "label"   # column name to use as y (required)

---

sklearn.train_test_split

Splits X and y into training and test sets.

Inputs: X, y

Outputs: X_train, X_test, y_train, y_test

- name: "split_data"
  component: "sklearn.train_test_split"
  parameters:
    test_size: 0.2
    random_state: 42

---

sklearn.<Estimator>.fit_predict

Trains any sklearn estimator on X and y, then predicts on the same data.

Inputs: X, y (use input_transform to map from upstream output names)

Outputs: model, predictions

Replace <Estimator> with any sklearn classifier or regressor class name:

- name: "train_model"
  component: "sklearn.LogisticRegression.fit_predict"
  input_transform:
    X: "X_train"
    y: "y_train"
  parameters:
    max_iter: 200
    C: 1.0

Other estimator examples:

• sklearn.RandomForestClassifier.fit_predict
• sklearn.MLPClassifier.fit_predict
• sklearn.GradientBoostingClassifier.fit_predict
• sklearn.SVC.fit_predict

All constructor keyword arguments can be passed under parameters.

---

sklearn.<Estimator>.predict

Runs inference with a previously trained model.

Inputs: model, X (use input_transform to map from upstream output names)

Outputs: predictions

- name: "predict_model"
  component: "sklearn.LogisticRegression.predict"
  input_transform:
    X: "X_test"
  output_mapping:
    predictions: "y_pred"
  metrics:
    - name: "accuracy"
      fn: "accuracy_score"
      inputs:
        y_true: "y_test"
        y_pred: "y_pred"

---

Metrics

The metrics block logs evaluation scores at the end of a process step. Each entry calls a scoring function from sklearn.metrics (or any callable accessible by name):

metrics:
  - name: "accuracy"
    fn: "accuracy_score"
    inputs:
      y_true: "y_train"     # key in process outputs
      y_pred: "predictions" # key in process outputs

Logged metrics are stored in the KV backend and can be queried via probe paths for use in chart components or custom @chart() functions.

---

Iterative Logging

For estimators that train iteratively (e.g. MLPClassifier), use iterative_logging to record a metric value at each training epoch:

- name: "train_mlp"
  component: "sklearn.MLPClassifier.fit_predict"
  input_transform:
    X: "X_train"
    y: "y_train"
  parameters:
    max_iter: 200
  iterative_logging:
    metric_name: "train_loss"   # name to store under in KV
    eval_split: "train"         # which split to evaluate on each iteration
  metrics:
    - name: "accuracy"
      fn: "accuracy_score"
      inputs:
        y_true: "y_train"
        y_pred: "predictions"

The logged values are stored step-by-step as {"1": val1, "2": val2, ...} and can be plotted as a time series using a matplotlib Histogram or a dynamic JS chart.

---

Full Example

The following config trains a LogisticRegression and an MLPClassifier in parallel, then evaluates both on the test set:

experiment:
  cache:
    backend:
      type: local
  pipeline:
    process_adjlist: |
      data_source split_data
      split_data train_model
      split_data predict_model
      train_model predict_model
      split_data train_mlp
      split_data predict_mlp
      train_mlp predict_mlp
    processes:
      - name: "data_source"
        component: "sklearn.csv_to_xy"
        parameters:
          label_column: "label"

      - name: "split_data"
        component: "sklearn.train_test_split"
        parameters:
          test_size: 0.2
          random_state: 42

      - name: "train_model"
        component: "sklearn.LogisticRegression.fit_predict"
        input_transform:
          X: "X_train"
          y: "y_train"
        parameters:
          max_iter: 200
        metrics:
          - name: "accuracy"
            fn: "accuracy_score"
            inputs:
              y_true: "y_train"
              y_pred: "predictions"

      - name: "predict_model"
        component: "sklearn.LogisticRegression.predict"
        input_transform:
          X: "X_test"
        output_mapping:
          predictions: "y_pred"
        metrics:
          - name: "accuracy"
            fn: "accuracy_score"
            inputs:
              y_true: "y_test"
              y_pred: "y_pred"

      - name: "train_mlp"
        component: "sklearn.MLPClassifier.fit_predict"
        input_transform:
          X: "X_train"
          y: "y_train"
        parameters:
          max_iter: 200
        iterative_logging:
          metric_name: "train_loss"
          eval_split: "train"
        metrics:
          - name: "accuracy"
            fn: "accuracy_score"
            inputs:
              y_true: "y_train"
              y_pred: "predictions"

      - name: "predict_mlp"
        component: "sklearn.MLPClassifier.predict"
        input_transform:
          X: "X_test"
        metrics:
          - name: "accuracy"
            fn: "accuracy_score"
            inputs:
              y_true: "y_test"
              y_pred: "predictions"

See also: matplotlib components for adding accuracy bar charts and loss histograms to this pipeline.