def _collect_lockdown_tooltip_layers(self, df, layers, base, cursor):
# Since it's difficult to support a condition of type 'mouseover OR checkbox'
# (due to broken alt.condition not two non-constants for if_true and if_false
# and due to completely broken checkboxes in Vega Lite), we'll add two layers
# to accompish the same thing. The first layer will be active when 'mouseover
# AND NOT checkbox' and the second layer will be active when just 'checkbox'.
def _make_base(tooltip):
ret = base.mark_text(
align='left',
dx=15,
dy=0,
font=self._font,
fontWeight=600,
).encode(
x=self._get_x(),
y=self._get_y(),
text=tooltip,
color=alt.value('black')
).transform_calculate(
# AGP lockdown_tooltip_text=f'datum.{self._detailby} + " " + datum.lockdown_type+ " " +"("+ datum.lockdown_date + ")"'
# lockdown_tooltip_text=f'datum.lockdown_type+ " " +"("+ datum.lockdown_date + ")"'
lockdown_tooltip_text=
'datum.lockdown_type + " " + "(" + datum.lockdown_date + ")"')
if 'event_index' in df.columns:
ret = ret.transform_filter('datum.event_index == 0')
return ret
lockdown_tooltip_text = 'lockdown_tooltip_text'
empty_if_checkbox = 'empty_if_checkbox'
hover_layer = _make_base(
alt.condition(cursor, f'{empty_if_checkbox}:N', alt.value(' ')))
# Checkbox to print event details
event_select = None
calculate_kwargs = {
empty_if_checkbox: f'datum.{lockdown_tooltip_text}'
}
if self.get('event_select', False):
event_checkbox = alt.binding_checkbox(
name='Show all intervention details ')
event_select = alt.selection_single(bind=event_checkbox,
name='events',
init={'values': False})
calculate_kwargs[
empty_if_checkbox] = f'{self._show_events()} ? " " : datum.{lockdown_tooltip_text}'
hover_layer = hover_layer.add_selection(event_select)
if 'Coverage' in df.columns:
# TODO: put this in a better place
# I'm just putting it here so that it appears earlier than the "show all events" checkbox for now
regional_checkbox = alt.binding_checkbox(
name='Hide regional intervention icons ')
regional_select = alt.selection_single(bind=regional_checkbox,
name='hide_regional_icons',
init={'values': False})
hover_layer = hover_layer.add_selection(regional_select)
hover_layer = hover_layer.transform_calculate(**calculate_kwargs)
layers['hover_layer'] = hover_layer
if event_select is not None:
layers['event_checkbox_layer'] = _make_base(
f'{lockdown_tooltip_text}:N').transform_filter(
self._show_events())
filter_genres = base.add_selection(genre_select).transform_filter(
genre_select).properties(title="Dropdown Filtering")
#color changing marks
rating_radio = alt.binding_radio(options=ratings)
rating_select = alt.selection_single(fields=['MPAA_Rating'],
bind=rating_radio,
name="Rating")
rating_color_condition = alt.condition(rating_select,
alt.Color('MPAA_Rating:N', legend=None),
alt.value('lightgray'))
highlight_ratings = base.add_selection(rating_select).encode(
color=rating_color_condition).properties(title="Radio Button Highlighting")
# Boolean selection for format changes
input_checkbox = alt.binding_checkbox()
checkbox_selection = alt.selection_single(bind=input_checkbox,
name="Big Budget Films")
size_checkbox_condition = alt.condition(
checkbox_selection, alt.SizeValue(25),
alt.Size('Hundred_Million_Production:Q'))
budget_sizing = base.add_selection(checkbox_selection).encode(
size=size_checkbox_condition).properties(title="Checkbox Formatting")
(filter_year | filter_genres) & (highlight_ratings | budget_sizing).show()
r = pdk.Deck(
layers=[layer],
initial_view_state=view_state,
mapbox_key=
"pk.eyJ1IjoieXVhbnl1YTQiLCJhIjoiY2tnaDN2ODNuMHFjdTM3cXQ1cjZ4ZjZ1bSJ9.7Rc5ptwp4PBKWFHC8iwPWg",
map_style="mapbox://styles/mapbox/light-v9")
st.sidebar.write("## NYC map for reference")
st.sidebar.write(
"If you find a location on the main plots that interest you, you can use the map below to dive in for more details."
)
st.sidebar.write(r)
with st.spinner(text="Loading..."):
selector_pick = alt.selection_interval(empty='none', encodings=['x', 'y'])
selector_drop = alt.selection_interval(empty='none', encodings=['x', 'y'])
binding_weekend = alt.binding_checkbox(name="weekend")
weekend = alt.selection_single(name=None,
fields=['weekend'],
bind=binding_weekend,
init={'weekend': True})
binding_weekday = alt.binding_checkbox(name="weekday")
weekday = alt.selection_single(name=None,
fields=['weekday'],
bind=binding_weekday,
init={'weekday': True})
taxi_map_dropoff_chart = alt.Chart(
taxi_map_pickup_dropoff,
title="Taxi Dropoffs (selectable)").mark_square(size=5).encode(
alt.Y('dropoff_latitude',
title='dropoff latitude',
df = df_raw
for variable in df_raw['variable'].unique():
this_df = df_raw[df_raw['variable'] == variable].copy()
this_df['value'] = list(np.diff(this_df['value']))+[0]
this_df['operation'] = ['diff']*n_row
df_raw = df_raw.append(this_df)
this_df['value'] = np.cumsum(this_df['value'])
this_df['operation'] = ['cumulative']*n_row
df_raw = df_raw.append(this_df)
line = alt.Chart(df).mark_line().encode(
x='sample',
y=alt.Y('value')
).properties(title='time series plots').interactive()
op_selections = [alt.selection_single(bind=alt.binding_checkbox(),
name=x, fields=['operation'])
for x in df_raw['operation'].unique()]
for op_selection in op_selections[::-1]:
line = line.add_selection(op_selection).transform_filter(op_selection)
dropdown_box = alt.binding_select(options=dataset_names)
dropdown_selection = alt.selection_single(bind=dropdown_box, name='dataset', fields=['variable'])
line = line.add_selection(dropdown_selection).transform_filter(dropdown_selection)
alt.vconcat(line,
data=df,
title="Altair Example of time series"
).save('altair_example.html')
def compile(self, df):
self.validate(df)
self[self.TRANSIENT] = DotDict()
try:
self._populate_transient_props(df)
base = alt.Chart(df, width=self._width, height=self._height)
layers = {}
extra_click_selection_kwargs = {}
if not self._manual_legend:
dropdown_options = [self.EMPTY_SELECTION]
dropdown_name = " "
if self.get('click_selection', False):
dropdown_options.extend(list(df[self._detailby].unique()))
dropdown_name = f'Select {self.get("readable_group_name", self.get("detailby", "group"))}: '
dropdown = alt.binding_select(options=dropdown_options,
name=dropdown_name)
extra_click_selection_kwargs['bind'] = dropdown
click_init = self.get('click_selection_init', None)
if click_init is not None:
click_init = {self._detailby: click_init}
if self._manual_legend:
click_init = [click_init]
extra_click_selection_kwargs['init'] = click_init
if self._manual_legend:
selection_type = getattr(alt, 'selection_multi')
else:
selection_type = getattr(alt, 'selection_single')
click_selection = selection_type(fields=[self._detailby],
on='click',
name=self.click,
empty='all',
clear='dblclick',
**extra_click_selection_kwargs)
# put a fake layer in first with no click selection
# since it has X and Y, it will help chart.interactive() to find x and y fields to bind to,
# allowing us to pan up and down and zoom over both axes instead of just 1.
layers['fake_interactive'] = base.mark_line().encode(
x=self._get_x(), y=self._get_y()).transform_filter('false')
# The first layer with a specified color channel is used to generate the legend.
# as such, we need to make sure the marks in the first layer w/ specified color channel are not translucent,
# otherwise we'll get a blank legend.
# We put a fake layer in here before we add any other layers w/ color channel specified, and we
# furthermore add the legend selection to it b/c it also seems like a multi-selection bound to the
# legend needs to be added to the layer that generates the legend.
if not self._manual_legend:
legend_selection = alt.selection_multi(
fields=[self._colorby],
on='click',
name=self.legend,
empty='all',
bind='legend',
clear='dblclick',
)
layers['legend'] = base.mark_point(size=0).encode(
x=self._get_x(), y=self._get_y(),
color=self._alt_color).add_selection(legend_selection)
extra_mouseover_kwargs = {}
if self._manual_legend:
# don't set clear for selection_single click --
# it will disappear when mouse is over points
extra_mouseover_kwargs['clear'] = 'mouseout'
cursor = alt.selection_single(name=self.cursor,
nearest=True,
on='mouseover',
fields=['x'],
empty='none',
**extra_mouseover_kwargs)
# Next goes the mouseover interaction layer (needs to happen before click selection layer,
# but only if click selection is "single" as opposed to "multi").
# We are binding the hover interaction to the 'x' column in the dataframe, so any other layers
# that make use of the hover interaction need to use this column for their x encoding in Altair.
# This means that things like lockdown tooltips need to use 'x' and then apply a filter to filter
# out non-lockdown days if they want work with the mouseover interaction.
if not self._manual_legend:
layers['selectors'] = base.mark_point(size=0).encode(
x=self._get_x()).add_selection(cursor)
# Put a fake layer in first to attach the click selection to. We use a fake layer for a few reasons.
# 1. It's not used as a base layer, so we won't get errors
# about the spec having the selection added multiple times.
# 2. We need a layer that has nice fat points that are easy to click for adding the click_selection to.
# Since the layer has fat points, however, the points need to be translucent, and since they
# need to be translucent, this layer cannot be the first layer that specifies color channel
# and similarly cannot be the layer with the legend_selection added on (in the case of using
# the built-in legend from Altair, which is only true if `use_manual_legend` is False).
layers['fake_points'] = self._make_point_layer(
base, point_size=400,
is_fake=True).add_selection(click_selection)
# Mouseover interaction goes after click interaction layer for manual legend, since manual legend
# uses `selection_multi` for the click layer which does not stick if going later (and does not appear
# at all if it goes first).
if self._manual_legend:
layers['selectors'] = self._make_point_layer(
base, point_size=0, is_fake=True).add_selection(cursor)
# now the meaty layers with actual content
layers['lines'] = self._make_line_layer(base)
layers['points'] = self._make_point_layer(
base,
point_size=self.get('point_size', self.DEFAULT_POINT_SIZE),
is_fake=False)
self._collect_tooltip_layers(df, layers, base, cursor)
if self.get('lockdown_extrapolation', True):
trend_checkbox = alt.binding_checkbox(
name='Show trend lines for selected ')
trend_select = alt.selection_single(bind=trend_checkbox,
name='trends',
init={'values': False})
layers[
'model_lines'] = self._make_lockdown_extrapolation_layer(
base)
layers[
'model_tooltip'] = self._make_extrapolation_tooltip_layer(
layers['model_lines'], cursor, trend_select)
if self.get('emoji_legend', False):
self._collect_emoji_legend_layers(df, layers)
layered = alt.layer(*layers.values())
layered = self._maybe_add_facet(layered)
if self.get('interactive', False):
layered = layered.interactive(bind_x=True, bind_y=True)
if self.get('title', False):
layered.title = self.get('title')
final_chart = layered
if self._manual_legend:
final_chart = alt.hconcat(final_chart,
self._make_manual_legend(
df, click_selection),
spacing=0)
final_chart = final_chart.configure(background=self.get(
'background', self.DEFAULT_BACKGROUND_COLOR))
final_chart = final_chart.configure_axis(titleFontSize=self.get(
'axes_title_fontsize', self.DEFAULT_AXES_TITLE_FONTSIZE), )
final_chart = final_chart.configure_title(font=self._font)
if not self._manual_legend:
final_chart = final_chart.configure_legend(
symbolType='diamond')
alt.themes.register('customFont', _fontSettings(self._font))
alt.themes.enable('customFont')
return final_chart
finally:
del self[self.TRANSIENT]
def compile(self, df):
self.validate(df)
self[self.TRANSIENT] = DotDict()
try:
self._populate_transient_props(df)
base = alt.Chart(df,
width=self.get('width', self.DEFAULT_WIDTH),
height=self.get('height', self.DEFAULT_HEIGHT))
layers = {}
dropdown_options = [self.EMPTY_SELECTION]
dropdown_name = " "
if self.get('click_selection', False):
dropdown_options.extend(list(df[self._detailby].unique()))
dropdown_name = f'Select {self.get("readable_group_name", self.detailby)}: '
dropdown = alt.binding_select(options=dropdown_options,
name=dropdown_name)
extra_click_selection_kwargs = {}
click_init = self.get('click_selection_init', None)
if click_init is not None:
extra_click_selection_kwargs['init'] = {
self._detailby: click_init
}
click_selection = alt.selection_single(
fields=[self._detailby],
on='click',
name=self.click,
empty='all',
bind=dropdown,
clear='dblclick',
**extra_click_selection_kwargs)
legend_selection = alt.selection_multi(
fields=[self._colorby],
on='click',
name=self.legend,
empty='all',
bind='legend',
clear='dblclick',
)
# put a fake layer in first with no click selection
# since it has X and Y, it will help chart.interactive() to find x and y fields to bind to,
# allowing us to pan up and down and zoom over both axes instead of just 1.
layers['fake_interactive'] = base.mark_line().encode(
x=self._get_x(), y=self._get_y()).transform_filter('false')
# next goes the tooltip selector layer (needs to happen before click selection layer)
cursor, selectors = self._make_cursor_selection(base)
layers['selectors'] = selectors
# The first layer with a specified color channel is used to generate the legend.
# as such, we need to make sure the marks in the first layer w/ specified color channel are not translucent,
# otherwise we'll get a blank legend.
# We put a fake layer in here before we add any other layers w/ color channel specified, and we
# furthermore add the legend selection to it b/c it also seems like a multi-selection bound to the
# legend needs to be added to the layer that generates the legend.
layers['legend'] = base.mark_point(size=0).encode(
x=self._get_x(), y=self._get_y(),
color=self._alt_color).add_selection(legend_selection)
# Put a fake layer in first to attach the click selection to. We use a fake layer for this for a few reasons.
# 1. It's not used as a base layer, so we won't get errors
# about the spec having the selection added multiple times.
# 2. We need a layer that has nice fat points that are easy to click for adding the click_selection to.
# Since the layer has fat points, however, the points need to be translucent, and since they
# need to be translucent, this layer cannot be the first layer that specifies color channel
# and similarly cannot be the layer with the legend_selection added on.
layers['fake_points'] = self._make_point_layer(
base, point_size=400,
is_fake=True).add_selection(click_selection)
# now the meaty layers with actual content
layers['lines'] = self._make_line_layer(base)
layers['points'] = self._make_point_layer(
base,
point_size=self.get('point_size', self.DEFAULT_POINT_SIZE),
is_fake=False)
self._collect_tooltip_layers(layers, base, cursor)
if self.get('lockdown_extrapolation', False):
trend_checkbox = alt.binding_checkbox(
name='Show trend lines for selected ')
trend_select = alt.selection_single(bind=trend_checkbox,
name='trends',
init={'values': True})
layers[
'model_lines'] = self._make_lockdown_extrapolation_layer(
base, trend_select)
layers[
'model_tooltip'] = self._make_extrapolation_tooltip_layer(
layers['model_lines'], cursor, trend_select)
layered = alt.layer(*layers.values())
layered = self._maybe_add_facet(layered)
if self.get('interactive', False):
layered = layered.interactive(bind_x=True, bind_y=True)
if self.get('title', False):
layered.title = self.get('title')
layered = layered.configure(background=self.get(
'background', self.DEFAULT_BACKGROUND_COLOR))
layered = layered.configure_axis(titleFontSize=self.get(
'axes_title_fontsize', self.DEFAULT_AXES_TITLE_FONTSIZE))
layered = layered.configure_legend(symbolType='diamond')
alt.themes.register('customFont', _fontSettings(self._font))
alt.themes.enable('customFont')
return layered
finally:
del self[self.TRANSIENT]
def rooflines(neural_network: str) -> alt.vegalite.v4.api.Chart:
"""
This function creates an Altair line chart with checkboxes. Creates a lot of them and then sums them up.
Note: Checkbox with default does not work after first click. Please keep track of when this issue is solved with this link: https://github.com/vega/vega-lite/issues/4870
Parameters
----------
data: pd.DataFrame
Dataframe from which the plot will be created.
neural_network:str
neural network that will also be plotted besides all hardware platforms.
Eg.:'imagenet|mnist|cifar'
'imagenet'
Returns
-------
Line Chart with checkboxes, all charts are summed up.
"""
maxX = 160000
width = 700
height = 500
nn_df, hw_df = clean_csv_rooflines(
path_topologies='data/cnn_topologies_compute_memory_requirements.csv',
path_hardware='data/peakPerfBandHardPlatf.csv')
#select data to be plotted according to user input
if neural_network in 'imagenet':
nn_df = nn_df[nn_df['Name'].str.contains(
"GoogLeNetv1|MobileNetv1|ResNet|EfficientNet")]
elif neural_network in 'cifar':
nn_df = nn_df[nn_df['Name'].str.contains("CNV")]
elif neural_network in 'mnist':
nn_df = nn_df[nn_df['Name'].str.contains("MLP")]
elif neural_network != 'imagenet|mnist|cifar':
return 'There were no results for the neural network asked. Please insert another network'
#regex expression replace for all MLPs because they overlapp in the plot because they have the same value for MLP100%,MLP50%..
nn_df = util.replace_data_df(nn_df, 'Name', [('MLP 100%', 'MLP*'),
('MLP 50%', 'MLP*'),
('MLP 25%', 'MLP*'),
('MLP 12.5%', 'MLP*')])
nn_df = nn_df.drop_duplicates()
#---------------Next part is to create all plots binded to checkboes-------------
#Selecting data for each checkbox, from dataset. Each checkbox will be tied to each one of these data
FPGA_data = hw_df[hw_df['Name'].str.contains("Ultra96 DPU|ZCU")]
NVIDIA_data = hw_df[hw_df['Name'].str.contains("TX2")]
GOOGLE_data = hw_df[hw_df['Name'].str.contains("EdgeTPU")]
INTEL_data = hw_df[hw_df['Name'].str.contains("NCS")]
INT2_data = nn_df[nn_df['Name'].str.contains("INT2")]
INT4_data = nn_df[nn_df['Name'].str.contains("INT4")]
INT8_data = nn_df[nn_df['Name'].str.contains("INT8")]
FP16_data = nn_df[nn_df['Name'].str.contains("FP16")]
FP32_data = nn_df[nn_df['Name'].str.contains("FP32")]
#Stating that the binding type will be a checkbox
filter_checkbox = alt.binding_checkbox()
#To create all checkboxes with the specifications info for each set
FPGA_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="FPGAs Ultra96 DPU ZCU ")
NVIDIA_select = alt.selection_single(
fields=["Hide"],
bind=filter_checkbox,
name="INVIDIA TX2 maxn, maxp, maxq ")
GOOGLE_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="GOOGLE EdgeTPU, fast, slow ")
INTEL_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="INTEL NCS ")
INT2_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="INT2")
INT4_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="INT4")
INT8_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="INT8")
FP16_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="FP16")
FP32_select = alt.selection_single(fields=["Hide"],
bind=filter_checkbox,
name="FP32")
#Color Conditions for each plot
FPGA_cond = alt.condition(FPGA_select, alt.Color("Name:N"),
alt.value("lightgray"))
NVIDIA_cond = alt.condition(NVIDIA_select, alt.Color("Name:N"),
alt.value("lightgray"))
GOOGLE_cond = alt.condition(GOOGLE_select, alt.Color("Name:N"),
alt.value("lightgray"))
INTEL_cond = alt.condition(INTEL_select, alt.Color("Name:N"),
alt.value("lightgray"))
INT2_cond = alt.condition(INT2_select, alt.Color("Name:N"),
alt.value("lightgray"))
INT4_cond = alt.condition(INT4_select, alt.Color("Name:N"),
alt.value("lightgray"))
INT8_cond = alt.condition(INT8_select, alt.Color("Name:N"),
alt.value("lightgray"))
FP16_cond = alt.condition(FP16_select, alt.Color("Name:N"),
alt.value("lightgray"))
FP32_cond = alt.condition(FP32_select, alt.Color("Name:N"),
alt.value("lightgray"))
#Creating all plots
FPGA_chart = line_chart_w_checkbox(FPGA_data, FPGA_cond, FPGA_select)
NVIDIA_chart = line_chart_w_checkbox(NVIDIA_data, NVIDIA_cond,
NVIDIA_select)
GOOGLE_chart = line_chart_w_checkbox(GOOGLE_data, GOOGLE_cond,
GOOGLE_select)
INTEL_chart = line_chart_w_checkbox(INTEL_data, INTEL_cond, INTEL_select)
INT2_chart = line_chart_w_checkbox(INT2_data, INT2_cond, INT2_select)
INT4_chart = line_chart_w_checkbox(INT4_data, INT4_cond, INT4_select)
INT8_chart = line_chart_w_checkbox(INT8_data, INT8_cond, INT8_select)
FP16_chart = line_chart_w_checkbox(FP16_data, FP16_cond, FP16_select)
FP32_chart = line_chart_w_checkbox(FP32_data, FP32_cond, FP32_select)
#--------------------------------------------------------------------------------------------------
# Create line plot
line_chart = alt.Chart().mark_line(clip=True).interactive().encode(
alt.X('arith_intens:Q'), alt.Y('performance:Q'),
alt.Color('Name:N', legend=alt.Legend(columns=2)))
#Create the selection which chooses nearest point on mouse hoover
selection = alt.selection(type='single',
nearest=True,
on='mouseover',
fields=['arith_intens'])
#to leave suggestions on, just replace arith_intens wiith anything else
#Create text plot to show the text values on mouse hoovering
text = (line_chart).mark_text(align='left', dx=3, dy=-3, clip=True).encode(
text=alt.condition(selection, 'Name:N', alt.value(' ')))
#Creates the points plot for the NNs. The points will be invisible
selectors = alt.Chart().add_selection(selection).mark_point(
clip=True).encode(
alt.X('arith_intens:Q'),
alt.Y('performance:Q'),
opacity=alt.value(0),
)
chart_all = (pd.Series(
[INT2_chart, INT4_chart, INT8_chart, FP16_chart, FP32_chart],
name="charts")).to_frame()
Chart = alt.layer(chart_all.charts.sum(numeric_only=False) + FPGA_chart +
NVIDIA_chart + GOOGLE_chart + INTEL_chart,
selectors,
text,
data=pd.concat([nn_df, hw_df]),
width=700,
height=500)
return Chart