google-research/timesfm · error · ValueError

train_dynamic_numerical_covariates and test_dynamic_numerica

Error message

train_dynamic_numerical_covariates and test_dynamic_numerical_covariates must be both present or both absent.

What it means

_assert_covariates enforces that dynamic numerical covariates are provided symmetrically: train_dynamic_numerical_covariates and test_dynamic_numerical_covariates must both be set or both be None. Supplying only one side raises this ValueError in create_covariate_matrix.

Source

Thrown at src/timesfm/utils/xreg_lib.py:221

      train_dynamic_categorical_covariates or {}
    )
    self.test_dynamic_numerical_covariates = test_dynamic_numerical_covariates or {}
    self.test_dynamic_categorical_covariates = test_dynamic_categorical_covariates or {}
    self.static_numerical_covariates = static_numerical_covariates or {}
    self.static_categorical_covariates = static_categorical_covariates or {}

  def _assert_covariates(self, assert_covariate_shapes: bool = False) -> None:
    """Verifies the validity of the covariate inputs."""

    # Check presence.
    if (
      self.train_dynamic_numerical_covariates
      and not self.test_dynamic_numerical_covariates
    ) or (
      not self.train_dynamic_numerical_covariates
      and self.test_dynamic_numerical_covariates
    ):
      raise ValueError(
        "train_dynamic_numerical_covariates and"
        " test_dynamic_numerical_covariates must be both present or both"
        " absent."
      )

    if (
      self.train_dynamic_categorical_covariates
      and not self.test_dynamic_categorical_covariates
    ) or (
      not self.train_dynamic_categorical_covariates
      and self.test_dynamic_categorical_covariates
    ):
      raise ValueError(
        "train_dynamic_categorical_covariates and"
        " test_dynamic_categorical_covariates must be both present or both"
        " absent."
      )

View on GitHub (pinned to 331c6d33cb)

Solutions

1. Provide both train_dynamic_numerical_covariates and test_dynamic_numerical_covariates with matching keys.
2. If no dynamic numerical covariates are intended, pass None (or omit) for both sides.
3. For future unknown values, supply placeholder/forecasted covariate values in the test dict.

Example fix

// before
forecast_with_covariates(inputs, dynamic_numerical_covariates=train_num, dynamic_categorical_covariates=None)
// after
forecast_with_covariates(inputs, dynamic_numerical_covariates=train_num, dynamic_categorical_covariates=None)  # ensure the API receives BOTH train_num and test_num dicts

assert (train_num is None) == (test_num is None), \
    "train and test dynamic numerical covariates must be both present or both absent"

def numerics_symmetric(train_num, test_num) -> bool:
    return (train_num is None) == (test_num is None)

try:
    forecaster.forecast_with_covariates(...)
except ValueError as e:
    if "dynamic_numerical_covariates must be both present" in str(e):
        test_num = {k: forecast_values for k in train_num}
        forecaster.forecast_with_covariates(...)
    else:
        raise

When it happens

Trigger: Calling forecast_with_covariates / create_covariate_matrix passing train_dynamic_numerical_covariates without test_dynamic_numerical_covariates (or vice versa).

Common situations: Forecasting future points where users assume test covariates can be omitted; providing historical-only covariates for model fitting; copying examples that only show one dict.

Related errors

• train_dynamic_categorical_covariates and test_dynamic_catego
• {dict_a_name} has keys not present in {dict_b_name}: {w}
• {dict_b_name} has keys not present in {dict_a_name}: {w}
• targets and train_lens must have the same number of elements
• At least one of dynamic_numerical_covariates, dynamic_catego

AI-assisted analysis of google-research/timesfm@331c6d33cb (2026-08-29). Data as JSON: /api/errors/2cef47b2b607590f. Report an issue: GitHub.