Skip to main content

Dataset Modeling

Prepare datasets for training by analyzing data health, features, and ensuring consistency for benchmarking.

Modeling Dataset

Dataset Health​

Dataset health provides insights into data quality and readiness:

Health Metrics:

  • Data Quality: Overall data quality score
  • Completeness: Percentage of non-null values
  • Consistency: Data consistency across features
  • Validity: Data validity and format compliance

Health Indicators:

  • Healthy: Dataset is in good condition
  • Warning: Some issues detected but usable
  • Critical: Significant issues that need attention

Health Checks:

  • Missing values
  • Data type consistency
  • Value ranges and distributions
  • Outlier detection
  • Data format validation

Feature Analysis​

Analyze features to understand their characteristics:

Analysis Types:

  • Statistical Analysis: Basic statistics for each feature
  • Distribution Analysis: Distribution of feature values
  • Correlation Analysis: Correlations between features
  • Importance Analysis: Feature importance for prediction

Analysis Results:

  • Feature Statistics: Mean, median, std dev, etc.
  • Value Distributions: Histograms and distributions
  • Correlation Matrix: Feature correlations
  • Missing Values: Missing value patterns
  • Outliers: Outlier detection and analysis

Using Analysis:

  • Feature Selection: Use analysis to select features
  • Data Quality: Identify data quality issues
  • Feature Engineering: Guide feature engineering decisions
  • Model Design: Inform model architecture decisions

Benchmark Consistency​

Ensure datasets remain consistent for benchmarking:

Consistency Requirements:

  • Fixed Schema: Schema should not change
  • Fixed Features: Feature set should remain constant
  • Fixed Partitioning: Partitioning should be reproducible
  • Version Control: Track dataset versions

Consistency Checks:

  • Schema validation
  • Feature set validation
  • Data distribution checks
  • Partition validation

Benchmark Datasets:

  • Use consistent datasets for fair model comparison
  • Maintain dataset versions for reproducibility
  • Document dataset characteristics
  • Track dataset changes

Preparing for Training​

Prepare datasets for training:

Preparation Steps:

  1. Complete Modeling: Finish feature and target selection
  2. Configure Partitions: Set up training/validation splits
  3. Review Health: Check dataset health metrics
  4. Validate Configuration: Ensure all requirements met
  5. Process Dataset: Process dataset to make it ready

Preparation Checklist:

  • ✅ Dataset structure selected
  • ✅ Features selected
  • ✅ Targets selected
  • ✅ Partitions configured
  • ✅ Dataset health acceptable
  • ✅ All requirements met

Ready for Training:

  • Dataset status is READY
  • All required configurations complete
  • Data processed and validated
  • Available for training jobs

Next Steps​