TSFileLoader

class torchtime.io.TSFileLoader(filepath: str, nan_replace_value: Union[int, float, str] = 'NaN')

File loader that can load time series files in sktimes .ts file format.

Parameters:

• filepath (str) – The path to the .ts file.

• nan_replace_value (int, float or str, optional) – The value, by which the missing value indicator “?” should be replaced. Default: “NaN”.

class State(value, names=_not_given, *values, module=None, qualname=None, type=None, start=1, boundary=None)

TSFileLoader’s internal parsing state.

as_tensor(return_targets: bool = False) → Union[Tensor, Tuple[Tensor, List[str]]]

Return the loaded data as a 3-dimensional tensor of the form (N, C, S).

Keyword Arguments:

return_targets (bool) –

Returns:

A 3 dimensional tensor.

Return type:

torch.Tensor

Parameters:

return_targets (bool) –

as_dataframe(return_targets: bool = False) → DataFrame

Return the loaded data as a pandas dataframe.

Keyword Arguments:

• return_targets (bool) – Identifies whether the targets should be included in the returned dataframe. If

• effect (True, the targets will be added as an additional column 'targets' to the dataframe. This only has an) –

• place. (if there are class labels available in the datafile that was parsed in the first) –

Returns:

A nested pandas dataframe holding the dimensions as columns and the number examples as rows, where every cell contains a pandas Series containing a univariate time series. If return_targets is set, it will also contain a column ‘targets’ that contains the class labels of every example.

Return type:

pd.DataFrame

Parameters:

return_targets (bool) –

get_classes()

Return the classes found in the ‘.ts’ file

Returns:

List of class names as string.

Return type:

List[str]

parse_header(line: str) → None

Parses a line of a .ts file header and updates the internal state of the loader with the extracted information.

Parameters:

line (str) – The header line to parse.

Returns:

None

Return type:

None

parse_header_value(value: str, value_type: TsTagValuePattern) → Union[bool, str, int, List[str]]

Parse a single header value that was extracted by the header line parser and return its value as the appropriate python object.

Parameters:

• value (str) – Extracted header value that should be parsed.

• value_type (TsTagValuePattern) – The expected type of the value, which should be applied.

Returns:

If the value is of type BOOLEAN. value converted to bool str: If the value is of type ANY_CONNECTED_STRING. Returns the stripped value string. List[str]: If the value is of type CLASS_LABEL. Returns a list of space separated string class labels.

Return type:

bool

parse_body(line: str) → None

Parse a line of the @data content of a .ts file if @timeStamps is False.

Parameters:

line (str) – The @data line to parse.

Returns:

None

Return type:

None

parse_timestamp_tuple(tuple_data: str, line_dim: int) → Tuple[Union[str, int], float]

Parse a timestamp tuple of the form (<timestamp>,<float>)

Parameters:

• tuple_data (str) – The timestamp tuple as string.

• line_dim (str) – The dimension of the line being current parsed.

Returns:

If the timestamp can be interpreted as an integer, return a tuple of the timestamp as int and the value as float. Tuple[str, float]: Else, the timestamp is probably a date, return a tuple of the timestamp as str and the value as float. The timestamp is later converted to an actual DateIndex using pandas.

Return type:

Tuple[int, float]

parse_body_timestamps(line: str) → None

Parse a line of the @data content of a .ts file if @timeStamps is True.

Parameters:

line (str) – The @data line to parse.

Returns:

None

Return type:

None

verify_metadata() → None

Verifies the parsed metadata of a .ts file by checking whether a full set of metadata has been provided.

Returns:

None

Return type:

None

parse()

Parses a .ts sktime formatted time series file.

Returns:

None