import altair as alt
import pandas as pd
from typing import Optional
from variationist.visualization.altair_chart import AltairChart
[docs]class TextOnlyBarChart(AltairChart):
"""A class for building a BarChart object for text-only computations."""
def __init__(
self,
df_data: pd.core.frame.DataFrame,
chart_metric: str,
metadata: dict,
extra_args: dict = {},
chart_dims: dict = {},
zoomable: Optional[bool] = True,
top_per_class_ngrams: Optional[int] = None,
) -> None:
"""
Initialization function for a building a BarChart object for text-only
computations.
Parameters
----------
df_data: pd.core.frame.DataFrame
A long-form dataframe storing the results of a prior analysis for a
given metric that will be used for visualization purposes.
chart_metric: str
The metric associated to the "df_data" dataframe and thus to the chart.
metadata: dict
A dictionary storing the metadata about the prior analysis.
extra_args: dict = {}
A dictionary storing the extra arguments for this chart type. Default = {}.
chart_dims: dict
The mapping dictionary for the variables for the given chart.
zoomable: Optional[bool] = True
Whether the (HTML) chart should be zoomable using the mouse or not (if this
is allowed for the resulting chart type by the underlying visualization
library).
top_per_class_ngrams: int = 20
The maximum number of highest scoring per-class n-grams to show (for bar
charts only). If set to None, it will show all the n-grams in the corpus
(it may easily be overwhelming). By default is 20 to keep the visualization
compact. This parameter is ignored when creating other chart types.
"""
super().__init__(df_data, chart_metric, metadata, extra_args, zoomable)
# @TODO: Eventually merge with standard bar chart
# Get relevant dimensions
x_name, x_type = "value", "quantitative"
y_name, y_type = "ngram", "nominal"
column_name, column_type = "text_name", "nominal"
color_name, color_type = "text_name", "nominal"
# Set attributes
self.top_per_class_ngrams = top_per_class_ngrams
self.metric_label = chart_metric + " value"
if self.n_cooc == 1:
self.text_label = (str(self.n_tokens) + "-gram") if self.n_tokens > 1 else "token"
else:
self.text_label = "tokens"
# Set base chart style
self.base_chart = self.base_chart.mark_bar(height=15, binSpacing=0.5, cornerRadiusEnd=5)
# Set dimensions
x_dim = alt.X(x_name, type=x_type, title=chart_metric)
y_dim = alt.Y(y_name, type=y_type, title="").sort("-x")
column_dim = alt.Column(column_name, type=column_type,
header=alt.Header(labelFontWeight="bold"))
color = alt.Color(color_name, color_type, legend=None) # for aestethics only
# Set tooltip
tooltip = [
alt.Tooltip(y_name, type=y_type, title=self.text_label),
alt.Tooltip(x_name, type=x_type, title=self.metric_label)
]
# Filter data to show up to k top ngrams (based on their value for the metric) for each group
self.base_chart = self.base_chart.transform_window(
rank = "rank(" + x_name + ")",
sort = [alt.SortField(x_name, order="descending"),
alt.SortField(y_name, order="ascending")], # break ties in ranking (@temp)
groupby = [column_name]
).transform_filter(
alt.datum.rank <= self.top_per_class_ngrams
)
# Encoding the data
self.base_chart = self.base_chart.encode(
x_dim,
y_dim,
column_dim,
color,
tooltip
)
# Set the independent dimensions
self.base_chart = self.base_chart.resolve_scale(
x="independent",
y="independent"
)
# Set extra properties
chart_width = max(100, 800 / len(list(df_data[column_name].unique())))
self.base_chart = self.base_chart.properties(width=chart_width, center=True)
# The chart has to be filterable, therefore create and add a search component to it
self.base_chart = self.add_search_component(self.base_chart, tooltip, y_dim)
# If the chart has to be zoomable, set the property (disallowed for bar chart)
# if self.zoomable == True:
# print(f"INFO: Zoom is disallowed for bar charts.")
# self.base_chart = self.base_chart.interactive()
# Create the final chart
self.chart = self.base_chart