def test_plot_multivariate():
"""Tests plot_multivariate function"""
x_col = "time"
df = pd.DataFrame({
x_col: [dt(2018, 1, 1),
dt(2018, 1, 2),
dt(2018, 1, 3)],
"oranges": [8.5, 2.0, 3.0],
"apples": [1.4, 2.1, 3.4],
"bananas": [4.2, 3.1, 3.0],
})
# plot with default values
fig = plot_multivariate(
df=df,
x_col=x_col)
assert fig.layout.showlegend
assert fig.layout.xaxis.title.text == x_col
assert fig.layout.yaxis.title.text == cst.VALUE_COL
assert len(fig.data) == 3
assert fig.data[0].mode == "lines"
assert fig.data[0].name == "oranges"
assert fig.data[1].name == "apples"
assert fig.data[2].name == "bananas"
# plot with style override
fig = plot_multivariate(
df=df,
x_col=x_col,
y_col_style_dict={
"oranges": {
"legendgroup": "one",
"line": {
"color": "red",
"dash": "dot"
}
},
"apples": None,
"bananas": {
"name": "plantain",
"legendgroup": "one",
"mode": "markers",
"line": None # Remove line params since we use mode="markers"
}
},
xlabel="xlab",
ylabel="ylab",
title="New Title",
showlegend=False
)
assert not fig.layout.showlegend
assert fig.layout.xaxis.title.text == "xlab"
assert fig.layout.yaxis.title.text == "ylab"
assert len(fig.data) == 3
assert fig.data[0].mode == "lines"
assert fig.data[0].legendgroup == "one"
assert fig.data[0].line.color == "red"
assert fig.data[0].name == "oranges"
assert fig.data[1].name == "apples"
assert fig.data[2].name == "plantain"
assert fig.data[2].mode == "markers"
# ylabel is used for default title
fig = plot_multivariate(
df=df,
x_col=x_col,
xlabel="xlab",
ylabel="ylab")
assert fig.layout.title.text == f"ylab vs xlab"
# plotly style
fig = plot_multivariate(
df=df,
x_col=x_col,
y_col_style_dict="plotly")
assert [fig.data[i].name for i in range(len(fig.data))] ==\
["oranges", "apples", "bananas"]
assert fig.data[0].line.color is None
assert fig.data[1].fill is None
# auto style
fig = plot_multivariate(
df=df,
x_col=x_col,
y_col_style_dict="auto")
assert [fig.data[i].name for i in range(len(fig.data))] ==\
["apples", "bananas", "oranges"] # sorted ascending
assert fig.data[0].line.color == "rgba(0, 145, 202, 1.0)"
assert fig.data[1].fill is None
# auto-fill style
fig = plot_multivariate(
df=df,
x_col=x_col,
y_col_style_dict="auto-fill",
default_color="blue")
assert [fig.data[i].name for i in range(len(fig.data))] ==\
["apples", "bananas", "oranges"] # sorted ascending
assert fig.data[0].line.color == "blue"
assert fig.data[1].fill == "tonexty"
assert fig.data[2].fill == "tonexty"
13, # accepts the same parameters as `plot_quantiles_and_overlays`
show_mean=True,
show_quantiles=False,
show_overlays=True,
center_values=
False, # note! does not center, to compute raw differences from the mean below
overlay_label_time_feature="str_dow",
)
overlay_minus_mean = grouped_df[OVERLAY_COL_GROUP] - grouped_df[
MEAN_COL_GROUP].values # subtracts the mean
x_col = overlay_minus_mean.index.name
overlay_minus_mean.reset_index(
inplace=True) # `plot_multivariate` expects the x-value to be a column
fig = plot_multivariate( # plots the deviation from the mean
df=overlay_minus_mean,
x_col=x_col,
ylabel=ts.original_value_col,
title="day of week effect over time")
plotly.io.show(fig)
# %%
# The pattern looks fairly stable until Nov 2013, when Monday
# far surpasses Sunday as the weekly peak. The relative values on Monday
# and Tuesday increase, and the relative values on Saturday and Sunday decline.
# Thus, it may be useful to include a seasonality changepoint around
# that time.
#
# .. tip::
# You can interact with the plot to focus on a particular day
# by double clicking its name in the legend. Double click again
# to unselect, or single click to show/hide a single series.
def plot_grouping_evaluation_metrics(
self,
metric_dict: Dict,
config_names: List = None,
which: str = "train",
groupby_time_feature: str = None,
groupby_sliding_window_size: int = None,
groupby_custom_column: pd.Series = None,
xlabel=None,
ylabel="Metric value",
title=None,
showlegend=True):
"""Returns a line plot of the grouped evaluation values of ``metric_dict`` of ``config_names``.
These values are grouped by the grouping method chosen by ``groupby_time_feature``,
``groupby_sliding_window_size`` and ``groupby_custom_column``.
See `~greykite.framework.output.univariate_forecast.UnivariateForecast.get_grouping_evaluation`
for details on grouping method.
Parameters
----------
metric_dict : `dict` [`str`, `callable`]
Evaluation metrics to compute. Same as
`~greykite.framework.framework.benchmark.benchmark_class.BenchmarkForecastConfig.get_evaluation_metrics`.
To get the best visualization, keep number of metrics <= 2.
config_names : `list` [`str`], default None
Which config results to plot. A list of config names.
If None, uses all the available config keys.
which: `str`
"train" or "test". Which dataset to evaluate.
groupby_time_feature : `str` or None, optional
If provided, groups by a column generated by
`~greykite.common.features.timeseries_features.build_time_features_df`.
See that function for valid values.
groupby_sliding_window_size : `int` or None, optional
If provided, sequentially partitions data into groups of size
``groupby_sliding_window_size``.
groupby_custom_column : `pandas.Series` or None, optional
If provided, groups by this column value. Should be same length as the DataFrame.
xlabel : `str` or None, default None
x-axis label. If None, label is determined by the groupby column name.
ylabel : `str` or None, default "Metric value"
y-axis label.
title : `str` or None, default None
Plot title. If None, default is based on ``config_name``.
showlegend : `bool`, default True
Whether to show the legend.
Returns
-------
fig : `plotly.graph_objs.Figure`
Interactive plotly graph.
"""
grouped_evaluation_df = self.get_grouping_evaluation_metrics(
metric_dict=metric_dict,
config_names=config_names,
which=which,
groupby_time_feature=groupby_time_feature,
groupby_sliding_window_size=groupby_sliding_window_size,
groupby_custom_column=groupby_custom_column)
# Figures out groupby_col name by process of elimination
cols = [
col for col in grouped_evaluation_df.columns
if col not in ["config_name", "split_num"]
]
groupby_col = get_pattern_cols(cols,
pos_pattern=".*",
neg_pattern=which)[0]
plot_df = (
grouped_evaluation_df.drop(columns=["split_num"
]) # Drops redundant column
.groupby(["config_name",
groupby_col]) # Averages values across splits
.mean().dropna(how="all") # Drops rows with all NA values
.unstack(
level=0
) # Moves config_name from multiindex rows to multiindex columns
.sort_index(
axis=1) # Sorts on groupby_col to plot groups in logical order
)
# Flattens and renames multiindex columns
cols = [groupby_col] + ["_".join(v) for v in plot_df.columns]
plot_df = pd.DataFrame(plot_df.to_records())
plot_df.columns = cols
if xlabel is None:
xlabel = groupby_col
if title is None:
title = f"{which} performance by {xlabel} across rolling windows"
fig = plot_multivariate(df=plot_df,
x_col=groupby_col,
y_col_style_dict="plotly",
xlabel=xlabel,
ylabel=ylabel,
title=title,
showlegend=showlegend)
return fig
def plot_flexible_grouping_evaluation(
self,
which="train",
groupby_time_feature=None,
groupby_sliding_window_size=None,
groupby_custom_column=None,
map_func_dict=None,
agg_kwargs=None,
extend_col_names=False,
y_col_style_dict="auto-fill",
default_color="rgba(0, 145, 202, 1.0)",
xlabel=None,
ylabel=None,
title=None,
showlegend=True):
"""Plots group-wise evaluation metrics. Whereas
`~greykite.framework.output.univariate_forecast.UnivariateForecast.plot_grouping_evaluation`
shows one metric, this can show any number of custom metrics.
For example:
* Mean and quantiles of squared error by group.
* Mean and quantiles of residuals by group.
* Mean and quantiles of actual and forecast by group.
* % of actuals outside prediction intervals by group
* any combination of the above metrics by the same group
See `~greykite.framework.output.univariate_forecast.UnivariateForecast.get_flexible_grouping_evaluation`
for details.
which: `str`
"train" or "test". Which dataset to evaluate.
groupby_time_feature : `str` or None, optional
If provided, groups by a column generated by
`~greykite.common.features.timeseries_features.build_time_features_df`.
See that function for valid values.
groupby_sliding_window_size : `int` or None, optional
If provided, sequentially partitions data into groups of size
``groupby_sliding_window_size``.
groupby_custom_column : `pandas.Series` or None, optional
If provided, groups by this column value. Should be same length as the DataFrame.
map_func_dict : `dict` [`str`, `callable`] or None, default None
Grouping evaluation metric specification, along with ``agg_kwargs``.
See `~greykite.framework.output.univariate_forecast.UnivariateForecast.get_flexible_grouping_evaluation`.
agg_kwargs : `dict` or None, default None
Grouping evaluation metric specification, along with ``map_func_dict``.
See `~greykite.framework.output.univariate_forecast.UnivariateForecast.get_flexible_grouping_evaluation`.
extend_col_names : `bool` or None, default False
How to name the grouping metrics.
See `~greykite.framework.output.univariate_forecast.UnivariateForecast.get_flexible_grouping_evaluation`.
y_col_style_dict: `dict` [`str`, `dict` or None] or "plotly" or "auto" or "auto-fill", default "auto-fill"
The column(s) to plot on the y-axis, and how to style them. The names should match
those generated by ``agg_kwargs`` and ``extend_col_names``.
The function
`~greykite.framework.output.univariate_forecast.UnivariateForecast.get_flexible_grouping_evaluation`
can be used to check the column names.
For convenience, start with "auto-fill" or "plotly", then adjust styling as needed.
See `~greykite.common.viz.timeseries_plotting.plot_multivariate` for details.
default_color: `str`, default "rgba(0, 145, 202, 1.0)" (blue)
Default line color when ``y_col_style_dict`` is one of "auto", "auto-fill".
xlabel : `str` or None, default None
x-axis label. If None, default is ``x_col``.
ylabel : `str` or None, default None
y-axis label. If None, y-axis is not labeled.
title : `str` or None, default None
Plot title. If None and ``ylabel`` is provided, a default title is used.
showlegend : `bool`, default True
Whether to show the legend.
Returns
-------
fig : `plotly.graph_objs.Figure`
Interactive plotly graph showing the evaluation metrics.
See `~greykite.common.viz.timeseries_plotting.plot_forecast_vs_actual`
return value for how to plot the figure and add customization.
See Also
--------
`~greykite.framework.output.univariate_forecast.UnivariateForecast.get_flexible_grouping_evaluation` : called by this function
`~greykite.common.viz.timeseries_plotting.plot_multivariate` : called by this function
"""
grouped_df = self.get_flexible_grouping_evaluation(
which=which,
groupby_time_feature=groupby_time_feature,
groupby_sliding_window_size=groupby_sliding_window_size,
groupby_custom_column=groupby_custom_column,
map_func_dict=map_func_dict,
agg_kwargs=agg_kwargs,
extend_col_names=extend_col_names)
x_col = grouped_df.index.name
grouped_df.reset_index(inplace=True)
fig = plot_multivariate(
grouped_df,
x_col=x_col,
y_col_style_dict=y_col_style_dict,
default_color=default_color,
xlabel=xlabel,
ylabel=ylabel,
title=title,
showlegend=showlegend)
return fig
def plot_forecasts_by_step(self,
forecast_step: int,
config_names: List = None,
xlabel: str = TIME_COL,
ylabel: str = VALUE_COL,
title: str = None,
showlegend: bool = True):
"""Returns a ``forecast_step`` ahead rolling forecast plot.
The plot consists one line for each valid. ``config_names``.
If available, the corresponding actual values are also plotted.
For a more customizable plot, see
:func:`~greykite.common.viz.timeseries_plotting.plot_multivariate`
Parameters
----------
forecast_step : `int`
Which forecast step to plot. A forecast step is an integer between 1 and the
forecast horizon, inclusive, indicating the number of periods from train end date
to the prediction date (# steps ahead).
config_names : `list` [`str`], default None
Which config results to plot. A list of config names.
If None, uses all the available config keys.
xlabel : `str` or None, default TIME_COL
x-axis label.
ylabel : `str` or None, default VALUE_COL
y-axis label.
title : `str` or None, default None
Plot title. If None, default is based on ``forecast_step``.
showlegend : `bool`, default True
Whether to show the legend.
Returns
-------
fig : `plotly.graph_objs.Figure`
Interactive plotly graph.
Plots multiple column(s) in ``self.forecasts`` against ``TIME_COL``.
See `~greykite.common.viz.timeseries_plotting.plot_forecast_vs_actual`
return value for how to plot the figure and add customization.
"""
if self.forecasts is None:
self.extract_forecasts()
if forecast_step > self.tscv.forecast_horizon:
raise ValueError(
f"`forecast_step` ({forecast_step}) must be less than or equal to "
f"forecast horizon ({self.tscv.forecast_horizon}).")
config_names = self.get_valid_config_names(config_names)
y_cols = [TIME_COL, ACTUAL_COL] + \
[f"{config_name}_{PREDICTED_COL}" for config_name in config_names]
df = self.forecasts[self.forecasts[FORECAST_STEP_COL] == forecast_step]
df = df[y_cols]
if title is None:
title = f"{forecast_step}-step ahead rolling forecasts"
fig = plot_multivariate(df=df,
x_col=TIME_COL,
y_col_style_dict="plotly",
xlabel=xlabel,
ylabel=ylabel,
title=title,
showlegend=showlegend)
return fig