def indicator_data(
indicators: IndicatorCollection) -> Tuple[List[np.ndarray], List[str]]:
r"""
Returns a tuple of a list of series and a list of names of series.
Each series, `S` is a timeseries of histograms of shape `[T, 10]`,
where `T` is the number of timesteps.
`S[:, 0]` is the `global_step`.
`S[:, 1:10]` represents the distribution at basis points
`0, 6.68, 15.87, 30.85, 50.00, 69.15, 84.13, 93.32, 100.00`.
Example:
>>> from labml import analytics
>>> indicators = analytics.runs('1d3f855874d811eabb9359457a24edc8')
>>> analytics.indicator_data(indicators)
"""
series, names = _cache.get_indicators_data(indicators)
if not series:
raise ValueError("No series found")
return series, names
def binned_heatmap(*args: any,
names: Optional[List[str]] = None,
x_name: Optional[str] = None,
height: int = 400,
width: int = 800,
height_minimap: int = 100):
r"""
Creates a scatter plot with Altair
This has multiple overloads
.. function:: binned_heatmap(indicators: IndicatorCollection, x_indicators: IndicatorCollection, *, names: Optional[List[str]] = None, x_name: Optional[str] = None, height: int = 400, width: int = 800, height_minimap: int = 100)
:noindex:
.. function:: binned_heatmap(series: List[np.ndarray], x_series: np.ndarray, *, names: Optional[List[str]] = None, x_name: Optional[str] = None, height: int = 400, width: int = 800, height_minimap: int = 100)
:noindex:
Arguments:
indicators(IndicatorCollection): Set of indicators to be plotted
x_indicators(IndicatorCollection): Indicator for x-axis
series(List[np.ndarray]): List of series of data
x_series(np.ndarray): X series of data
Keyword Arguments:
names(List[str]): List of names of series
name(str): Name of X series
noise: Noise to be added to spread out the scatter plot
circle_size: size of circles in the plot
height: height of the visualization
width: width of the visualization
height_minimap: height of the view finder
Return:
The Altair visualization
Example:
>>> from labml import analytics
>>> indicators = analytics.runs('1d3f855874d811eabb9359457a24edc8')
>>> analytics.scatter(indicators.validation_loss, indicators.train_loss)
"""
series = None
x_series = None
if len(args) == 2:
if isinstance(args[0], _IndicatorCollection) and isinstance(
args[1], _IndicatorCollection):
series, names_ = _cache.get_indicators_data(args[0])
x_series, x_name_ = _cache.get_indicators_data(args[1])
if len(x_series) != 1:
raise ValueError(
"There should be exactly one series for x-axis")
if not series:
raise ValueError("No series found")
x_series = x_series[0]
if x_name is None:
x_name = x_name_[0]
if names is None:
names = names_
elif isinstance(args[0], list):
series = args[0]
x_series = args[1]
if series is None:
raise ValueError(
"scatter should be called with an indicator collection"
" or a series. Check documentation for details.")
if x_name is None:
x_name = 'x'
if names is None:
names = [f'{i + 1}' for i in range(len(series))]
tables = [_binned_heatmap.data_to_table(s, x_series) for s in series]
names = _remove_names_prefix(names)
return _binned_heatmap.render(tables,
names=names,
x_name=x_name,
width=width,
height=height,
height_minimap=height_minimap)
def distribution(*args: any,
names: Optional[List[str]] = None,
levels: int = 5,
alpha: int = 0.6,
height: int = 400,
width: int = 800,
height_minimap: int = 100):
r"""
Creates a distribution plot distribution with Altair
This has multiple overloads
.. function:: distribution(indicators: IndicatorCollection, *, names: Optional[List[str]] = None, levels: int = 5, alpha: int = 0.6, height: int = 400, width: int = 800, height_minimap: int = 100)
:noindex:
.. function:: distribution(series: Union[np.ndarray, torch.Tensor], *, names: Optional[List[str]] = None, levels: int = 5, alpha: int = 0.6, height: int = 400, width: int = 800, height_minimap: int = 100)
:noindex:
.. function:: distribution(series: List[Union[np.ndarray, torch.Tensor]], *, names: Optional[List[str]] = None, levels: int = 5, alpha: int = 0.6, height: int = 400, width: int = 800, height_minimap: int = 100)
:noindex:
.. function:: distribution(series: List[Union[np.ndarray, torch.Tensor]], step: np.ndarray, *, names: Optional[List[str]] = None, levels: int = 5, alpha: int = 0.6, height: int = 400, width: int = 800, height_minimap: int = 100)
:noindex:
Arguments:
indicators(IndicatorCollection): Set of indicators to be plotted
series(List[np.ndarray]): List of series of data
step(np.ndarray): Steps
Keyword Arguments:
names(List[str]): List of names of series
levels: how many levels of the distribution to be plotted
alpha: opacity of the distribution
height: height of the visualization
width: width of the visualization
height_minimap: height of the view finder
Return:
The Altair visualization
Example:
>>> from labml import analytics
>>> indicators = analytics.runs('1d3f855874d811eabb9359457a24edc8')
>>> analytics.distribution(indicators)
"""
series = None
step = None
if len(args) == 1:
if isinstance(args[0], _IndicatorCollection):
series, names_ = _cache.get_indicators_data(args[0])
if not series:
raise ValueError("No series found")
if names is None:
names = names_
elif isinstance(args[0], list):
series = args[0]
else:
series = [args[0]]
elif len(args) == 2:
series = args[0]
step = args[1]
if names is None:
names = [f'{i + 1}' for i in range(len(series))]
if series is None:
raise ValueError(
"distribution should be called with an indicator collection"
" or a series. Check documentation for details.")
names = _remove_names_prefix(names)
tables = _density.data_to_table(series, names, step)
return _density.render(tables,
levels=levels,
alpha=alpha,
width=width,
height=height,
height_minimap=height_minimap)