---
title: RawData (latest version)
description: API reference for qiskit_noise_learning.data.RawData in the latest version of qiskit-noise-learning
source: https://quantum.cloud.ibm.com/docs/en/api/qiskit-noise-learning/generated/data-raw-data
---

# qiskit\_noise\_learning.data.RawData

*class* `qiskit_noise_learning.data.RawData(datatree: DataTree)`

[GitHub](https://github.com/Qiskit/qiskit-noise-learning/tree/stable/0.1/qiskit_noise_learning/data/raw_data.py)

Bases: [`LeveledData`](/docs/api/qiskit-noise-learning/generated/data-leveled-data "qiskit_noise_learning.data.leveled_data.LeveledData")

Raw experimental outcome data associated with instruction sequences and classical registers.

This class is a wrapper around a 1-layer deep XArray `DataTree` with arbitrary string keys. Each leaf dataset contains:

- Data variables:

  > - `data`: The raw boolean data with dimensions `("randomization", "shot", "bit")`.
  > - `data_mask`: A boolean mask with dimensions `("randomization", "shot")`. Handles potential raggedness in the `"shot"` dimension across different randomizations.
  > - `measurement_flips`: A boolean array of measurement flips with dimensions `("randomization", "bit")`.
  > - `time_lbs`: Lower bound on data acquisition times, with dimensions `("randomization",)`, of type `"datetime64[us]"`.
  > - `time_ubs`: Upper bound on data acquisition times, with dimensions `("randomization",)`, of type `"datetime64[us]"`.

- Coordinates:

  > - `unbound_instruction_sequence`: The unbound instruction sequence for the data, along dimension `("randomization",)`, of type `InstructionSequence`.
  > - `fragment_depth`: Integer array of fragment depths along dimension `("randomization",)`.

- Attrs:

  > - `creg_names`: Ordered list of classical register names.
  > - `measurement_map`: Dictionary mapping creg names to arrays of measured qubit indices.
  > - `creg_bit_boundaries`: Dictionary mapping creg names to `(start_idx, end_idx)` tuples indicating the slice of the `"bit"` dimension for that register.

Datasets are grouped by creg metadata: two datasets with the same `creg_names` and `measurement_map` are merged along the `"randomization"` dimension.

**Parameters**

**datatree** – A datatree in the above format.

### \_\_init\_\_

`__init__(datatree: DataTree)`

## Methods

|                                                                                                                                                       |                                                 |
| ----------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------- |
| [`__init__`](#qiskit_noise_learning.data.RawData.__init__ "qiskit_noise_learning.data.RawData.__init__")(datatree)                                    |                                                 |
| [`filter_time`](#qiskit_noise_learning.data.RawData.filter_time "qiskit_noise_learning.data.RawData.filter_time")(lb, ub)                             | Filter to data gathered within the time bounds. |
| [`from_arrays`](#qiskit_noise_learning.data.RawData.from_arrays "qiskit_noise_learning.data.RawData.from_arrays")(creg\_names, measurement\_map, ...) | Instantiate from data specified as arrays.      |
| [`merge`](#qiskit_noise_learning.data.RawData.merge "qiskit_noise_learning.data.RawData.merge")(other)                                                | Merge with another raw data set.                |

## Attributes

|                                                                                                          |                |
| -------------------------------------------------------------------------------------------------------- | -------------- |
| [`datatree`](#qiskit_noise_learning.data.RawData.datatree "qiskit_noise_learning.data.RawData.datatree") | The data tree. |

### datatree

Type: `DataTree`

The data tree.

### from\_arrays

*classmethod* `from_arrays(creg_names: list[str], measurement_map: dict[str, ndarray], instruction_sequences: list[InstructionSequence], data: list[ndarray[bool]], measurement_flips: list[ndarray[bool]], time_lbs: list[ndarray[datetime64]], time_ubs: list[ndarray[datetime64]])`

Instantiate from data specified as arrays.

All instruction sequences must share the same creg structure (same `creg_names` and `measurement_map`). The resulting `RawData` contains a single-leaf datatree.

**Parameters**

- **creg\_names** – Ordered list of classical register names.
- **measurement\_map** – Dictionary mapping creg names to arrays of measured physical qubit indices.
- **instruction\_sequences** – The list of instruction sequences used to generate the experiments.
- **data** – A list of outcome data for each instruction sequence for all classical registers. The data has dimensions `("randomization", "shot", "bit")`. Bits are ordered according to `creg_names` order, with each creg’s bits contiguous.
- **measurement\_flips** – A list of measurement flips to be applied to the data for each instruction sequence. Dimensions are `("randomization", "bit")`.
- **time\_lbs** – A lower bound on the data collection time for each randomization for a given instruction sequence. The dimensions are `("randomization",)`.
- **time\_ubs** – An upper bound on the data collection time for each randomization for a given instruction sequence. The dimensions are `("randomization",)`.

### merge

`merge(other: Self) → Self`

Merge with another raw data set.

Datasets with matching creg metadata (`creg_names` and `measurement_map`) are concatenated along the `"randomization"` dimension. Potential raggedness of the `"shot"` dimension is handled via the `"data_mask"` data variable.

**Parameters**

**other** – The other raw dataset.

**Returns**

The merged data.

### filter\_time

`filter_time(lb: datetime64, ub: datetime64) → Self`

Filter to data gathered within the time bounds.

**Parameters**

- **lb** – The time lower bound (inclusive).
- **ub** – The time upper bound (inclusive).

**Returns**

The time filtered version of self.