Join TensorFlow at Google I/O, May 11-12 Register now


Infers frequency and makes an irregular time series regular.

Converts a time series into a regular time series having the same period between successive time points (e.g. 5 seconds, or 1 day). If the frequency is known, it can be supplied through the 'frequency' argument; otherwise it will be inferred.

If multiple values share the same timestamp, they are summed into a single value.

series a Pandas pd.Series instance indexed by a pd.DateTimeIndex. This may also be a single-column pd.DataFrame.
frequency a Pandas DateOffset object, e.g. pd.DateOffset(seconds=1). If no frequency is specified, and the index of series does not have a frequency populated, the granularity of the time series will be inferred automatically. Default value: None.
warn_missing_tolerance optional instance of tfp.sts.MissingValuesTolerance, specifying warning thresholds for too many missing values. Default value: None. (do not warn).
err_missing_tolerance optional instance of tfp.sts.MissingValuesTolerance, specifying error thresholds for too many missing values. Default value: None. (do not raise errors).
max_series_length int maximum length of the regularized series (note that regularization may increase the length of the series). Used to bound the resources used per invocation. Default value: None.

regularized_series instance of the same type as series (pd.Series or pd.DataFrame) whose index follows a regular frequency (regularized_series.index.freq is not None). Any values not provided are filled in as NaN.

TypeError if data is not an instance of pd.Series or pd.DataFrame.
ValueError if data is empty, data.index is not a DatetimeIndex, data.index is not sorted, or if applying the inferred frequency would exceed the max_series_length or create more missing values than allowed by err_missing_vals.