google-research/timesfm · error · ValueError

targets and train_lens must have the same number of elements

Error message

targets and train_lens must have the same number of elements.

What it means

_assert_covariates also validates shape consistency: the number of targets (time series) must equal the number of train_lens entries. A mismatch means the covariate-holder was constructed with inconsistent-length lists, so the raise prevents building a broken regression matrix.

Source

Thrown at src/timesfm/utils/xreg_lib.py:263

        "train_dynamic_numerical_covariates",
        "test_dynamic_numerical_covariates",
      ),
      (
        self.train_dynamic_categorical_covariates,
        self.test_dynamic_categorical_covariates,
        "train_dynamic_categorical_covariates",
        "test_dynamic_categorical_covariates",
      ),
    ):
      if w := set(dict_a.keys()) - set(dict_b.keys()):
        raise ValueError(f"{dict_a_name} has keys not present in {dict_b_name}: {w}")
      if w := set(dict_b.keys()) - set(dict_a.keys()):
        raise ValueError(f"{dict_b_name} has keys not present in {dict_a_name}: {w}")

    # Check shapes.
    if assert_covariate_shapes:
      if len(self.targets) != len(self.train_lens):
        raise ValueError(
          "targets and train_lens must have the same number of elements."
        )

      if len(self.train_lens) != len(self.test_lens):
        raise ValueError(
          "train_lens and test_lens must have the same number of elements."
        )

      for i, (target, train_len) in enumerate(zip(self.targets, self.train_lens)):
        if len(target) != train_len:
          raise ValueError(
            f"targets[{i}] has length {len(target)} != expected {train_len}."
          )

      for key, values in self.static_numerical_covariates.items():
        if len(values) != len(self.train_lens):
          raise ValueError(
            f"static_numerical_covariates has key {key} with number of"

View on GitHub (pinned to 331c6d33cb)

Solutions

Rebuild the covariates object ensuring targets and train_lens are derived from the same filtered list of series.
Check lengths: assert len(targets) == len(train_lens) == len(test_lens) before calling create_covariate_matrix.
If series were filtered, apply the same filter to train_lens and all covariate dicts.

Example fix

// before
covs = make_covariates(targets=targets, train_lens=[len(t) for t in targets[:4]])  # len mismatch
// after
covs = make_covariates(targets=targets, train_lens=[len(t) for t in targets])

Defensive patterns

Strategy: validation

Validate before calling

assert len(targets) == len(train_lens) == len(test_lens), \
    f"targets({len(targets)}), train_lens({len(train_lens)}), test_lens({len(test_lens)}) length mismatch"

Type guard

def lengths_consistent(targets, train_lens, test_lens) -> bool:
    return len(targets) == len(train_lens) == len(test_lens)

Try / catch

try:
    covs.create_covariate_matrix()
except ValueError as e:
    if "must have the same number of elements" in str(e):
        print("Rebuild covariates object from the same filtered series list:", e)
        raise

Prevention

Derive targets, train_lens, and test_lens from one loop over the same series list
Filter series before constructing the covariates object, not after
Add a length-consistency assert in your data-prep code

When it happens

Trigger: create_covariate_matrix → _assert_covariates (when assert_covariate_shapes is true) where len(self.targets) != len(self.train_lens), e.g. an XRegCovariates-like object built with per-series covariates lists of different lengths than the targets list.

Common situations: Constructing the covariates object with targets from one dataset and train_lens computed from another (series dropped/added); filtering series but not the lens; off-by-one or duplicated entries.

Related errors

AI-assisted analysis of google-research/timesfm@331c6d33cb (2026-08-29). Data as JSON: /api/errors/9ad4e44e61810df3. Report an issue: GitHub.