def plot_parallel_coordinates(
plotter: Plotter,
mcmc_tables: List[pd.DataFrame],
mcmc_params: List[pd.DataFrame],
):
parameters = [
param for param in mcmc_params[0].loc[:, "name"].unique().tolist()
if "dispersion_param" not in param
]
target_n_lines = 500.0
n_samples = int(target_n_lines / len(mcmc_tables))
combined_mcmc_df = merge_and_pivot_mcmc_parameters_loglike(
mcmc_tables, mcmc_params, parameters, n_samples_per_chain=n_samples)
w = len(parameters) * 200
h = 800
labels = {}
for param in parameters:
labels[param] = get_plot_text_dict(param)
figure = px.parallel_coordinates(
combined_mcmc_df,
color="fitness",
dimensions=parameters,
labels=labels,
color_continuous_scale=px.colors.diverging.Tealrose,
height=h,
width=w,
)
figure.show()
def plot_mixing_matrix_2(plotter: Plotter, iso3: str):
fig, axes, _, n_rows, n_cols, _ = plotter.get_figure(n_panels=6)
fig.tight_layout()
positions = {
"all_locations": [0, 0],
"home": [0, 1],
"work": [0, 2],
"other_locations": [1, 1],
"school": [1, 2],
"none": [1, 0],
}
for location, position in positions.items():
axis = axes[position[0], position[1]]
if location != "none":
mixing_matrix = get_country_mixing_matrix(location, iso3)
im = axis.imshow(mixing_matrix, cmap="hot", interpolation="none", extent=[0, 80, 80, 0])
axis.set_title(get_plot_text_dict(location), fontsize=12)
axis.set_xticks([5, 25, 45, 65])
axis.set_yticks([5, 25, 45, 65])
axis.tick_params(top=True, bottom=False, labeltop=True, labelbottom=False)
cbar = axis.figure.colorbar(im, ax=axis, cmap="hot")
else:
axis.axis("off")
plotter.save_figure(fig, filename="mixing-matrix", title_text="Mixing matrix")
def plot_time_varying_multi_input(
plotter: Plotter,
tv_key: str,
times: List[float],
is_logscale: bool,
):
"""
Plot single simple plot of a function over time
"""
# Plot requested func names.
fig, axes, max_dims, n_rows, n_cols, _ = plotter.get_figure()
if is_logscale:
axes.set_yscale("log")
df = pd.DataFrame(tv_key)
df.index = times
axes.plot(df.index, df.values)
change_xaxis_to_date(axes, REF_DATE)
pyplot.legend(
df.columns, loc="best", labels=[get_plot_text_dict(location) for location in df.columns]
)
if X_MIN is not None and X_MAX is not None:
axes.set_xlim((X_MIN, X_MAX))
axes.set_ylim(bottom=0.0)
plotter.save_figure(
fig, filename=f"time-variant-{'Google mobility'}", title_text="Google mobility"
)
def plot_param_matrix(
plotter: StreamlitPlotter,
mcmc_params: List[pd.DataFrame],
parameters: List,
label_param_string=False,
show_ticks=False,
file_name="",
):
burn_in, label_font_size, label_chars, bins, style, dpi_request = BURN_INS, 8, 2, 20, "Shade", 300
plots.calibration.plots.plot_param_vs_param(
plotter,
mcmc_params,
parameters,
burn_in,
style,
bins,
label_font_size,
label_chars,
dpi_request,
label_param_string=label_param_string,
show_ticks=show_ticks,
file_name=file_name,
)
param_names = [get_plot_text_dict(param) for param in parameters]
params_df = pd.DataFrame({"names": param_names})
params_df["numbers"] = range(1, len(params_df) + 1)
create_downloadable_csv(params_df, "parameter_indices")
key_string = ""
for i_param, param_name in enumerate(param_names):
key_string += str(i_param + 1) + ", " + param_name + "; "
st.write(key_string)
st.dataframe(params_df)
def plot_multiple_posteriors(
plotter: Plotter,
mcmc_params: List[pd.DataFrame],
mcmc_tables: List[pd.DataFrame],
burn_in: int,
num_bins: int,
title_font_size: int,
label_font_size: int,
capitalise_first_letter: bool,
dpi_request: int,
priors: list,
parameters: list,
file_name="all_posteriors",
):
"""
Plots the posterior distribution of a given parameter in a histogram.
"""
# Except not the dispersion parameters - only the epidemiological ones
fig, axes, _, n_rows, n_cols, indices = plotter.get_figure(len(parameters))
for i in range(n_rows * n_cols):
axis = axes[indices[i][0], indices[i][1]]
if i < len(parameters):
param_name = parameters[i]
vals_df = get_posterior(mcmc_params, mcmc_tables, param_name,
burn_in)
for i, prior in enumerate(priors):
if prior["param_name"] == param_name:
prior = priors[i]
break
x_range = workout_plot_x_range(prior)
x_values = np.linspace(x_range[0], x_range[1], num=1000)
y_values = [calculate_prior(prior, x, log=False) for x in x_values]
# Plot histograms
vals_df.hist(bins=num_bins, ax=axis, density=True)
# Plot the prior
axis.plot(x_values, y_values)
axis.set_title(
get_plot_text_dict(
param_name,
capitalise_first_letter=capitalise_first_letter),
fontsize=title_font_size,
)
pyplot.setp(axis.get_yticklabels(), fontsize=label_font_size)
pyplot.setp(axis.get_xticklabels(), fontsize=label_font_size)
else:
axis.axis("off")
fig.tight_layout()
plotter.save_figure(fig, filename=file_name, dpi_request=dpi_request)
def plot_multiple_param_traces(
plotter: Plotter,
mcmc_params: List[pd.DataFrame],
burn_in: int,
title_font_size: int,
label_font_size: int,
capitalise_first_letter: bool,
dpi_request: int,
optional_param_request=None,
file_name="all_traces",
x_ticks_on=True,
):
# Except not the dispersion parameters - only the epidemiological ones
parameters = [
param for param in mcmc_params[0].loc[:, "name"].unique().tolist()
if "dispersion_param" not in param
]
params_to_plot = optional_param_request if optional_param_request else parameters
fig, axes, _, n_rows, n_cols, indices = plotter.get_figure(
len(params_to_plot), share_xaxis=True, share_yaxis=False)
for i in range(n_rows * n_cols):
axis = axes[indices[i][0], indices[i][1]]
if i < len(params_to_plot):
param_name = params_to_plot[i]
for i_chain in range(mcmc_params[0]["chain"].iloc[-1]):
param_mask = (mcmc_params[0]["chain"] == i_chain) & (
mcmc_params[0]["name"] == param_name)
param_values = mcmc_params[0][param_mask].values
axis.plot(param_values[:, 3], alpha=0.8, linewidth=0.7)
axis.set_title(
get_plot_text_dict(
param_name,
capitalise_first_letter=capitalise_first_letter),
fontsize=title_font_size,
)
if not x_ticks_on:
axis.set_xticks([])
elif indices[i][0] == n_rows - 1:
x_label = "Iterations" if capitalise_first_letter else "iterations"
axis.set_xlabel(x_label, fontsize=label_font_size)
pyplot.setp(axis.get_yticklabels(), fontsize=label_font_size)
pyplot.setp(axis.get_xticklabels(), fontsize=label_font_size)
if burn_in > 0:
axis.axvline(x=burn_in,
color=COLOR_THEME[1],
linestyle="dotted")
else:
axis.axis("off")
fig.tight_layout()
plotter.save_figure(fig, filename=file_name, dpi_request=dpi_request)
def plot_all_params_vs_loglike(
plotter: Plotter,
mcmc_tables: List[pd.DataFrame],
mcmc_params: List[pd.DataFrame],
burn_in: int,
title_font_size: int,
label_font_size: int,
capitalise_first_letter: bool,
dpi_request: int,
):
# Except not the dispersion parameters - only the epidemiological ones
parameters = [
param for param in mcmc_params[0].loc[:, "name"].unique().tolist()
if "dispersion_param" not in param
]
fig, axes, _, n_rows, n_cols, indices = plotter.get_figure(
len(parameters), share_xaxis=False, share_yaxis=True)
for i in range(n_rows * n_cols):
axis = axes[indices[i][0], indices[i][1]]
if i < len(parameters):
param_name = parameters[i]
plot_param_vs_loglike(mcmc_tables, mcmc_params, param_name,
burn_in, axis)
axis.set_title(
get_plot_text_dict(
param_name,
capitalise_first_letter=capitalise_first_letter),
fontsize=title_font_size,
)
if indices[i][0] == n_rows - 1:
x_label = "Iterations" if capitalise_first_letter else "iterations"
axis.set_xlabel(x_label, fontsize=label_font_size)
pyplot.setp(axis.get_yticklabels(), fontsize=label_font_size)
pyplot.setp(axis.get_xticklabels(), fontsize=label_font_size)
else:
axis.axis("off")
fig.tight_layout()
plotter.save_figure(fig,
filename=f"all_posteriors",
dpi_request=dpi_request)
def plot_multi_cdr_curves(plotter: Plotter, times, detected_proportions,
start_date, end_date, rotation, regions):
"""
Plot multiple sets of CDR curves onto a multi-panel figure
"""
fig, axes, _, n_rows, n_cols, indices = plotter.get_figure(
n_panels=len(regions))
# Loop over models and plot
for i_region in range(n_rows * n_cols):
axis = axes[indices[i_region][0], indices[i_region][1]]
if i_region < len(regions):
axis = plot_cdr_to_axis(axis, times,
detected_proportions[i_region])
tidy_cdr_axis(axis, rotation, start_date, end_date)
axis.set_title(get_plot_text_dict(regions[i_region]))
else:
axis.axis("off")
fig.tight_layout()
plotter.save_figure(fig, filename=f"multi_cdr_curves")
def plot_multi_fit(
plotter: Plotter,
output_names: list,
outputs: dict,
targets,
is_logscale=False,
title_font_size=8,
label_font_size=8,
dpi_request=300,
capitalise_first_letter=False,
):
fig, axes, _, n_rows, n_cols, indices = plotter.get_figure(
len(output_names), share_xaxis=True)
for i_output in range(n_rows * n_cols):
if i_output < len(output_names):
output = output_names[i_output]
axis = plot_calibration(
axes[indices[i_output][0], indices[i_output][1]],
output,
outputs[output],
targets,
is_logscale,
)
change_xaxis_to_date(axis, REF_DATE, rotation=0)
axis.set_title(
get_plot_text_dict(
output, capitalise_first_letter=capitalise_first_letter),
fontsize=title_font_size,
)
filename = f"calibration-fit-{output}"
else:
axes[indices[i_output][0], indices[i_output][1]].axis("off")
fig.tight_layout()
plotter.save_figure(fig, filename=filename, dpi_request=dpi_request)
def plot_vic_seroprevalences(
plotter: Plotter,
uncertainty_df: pd.DataFrame,
scenario_id: int,
time: float,
ref_date=REF_DATE,
name="",
requested_quantiles=None,
credible_range=50,
):
fig, axes, _, _, _, _ = plotter.get_figure(n_panels=2, share_yaxis="all")
cluster_axis, age_axis = axes
mask = (uncertainty_df["scenario"]
== scenario_id) & (uncertainty_df["time"] == time)
df = uncertainty_df[mask]
quantile_vals = df["quantile"].unique().tolist()
seroprevalence_by_cluster = {}
sero_outputs = [
output for output in df["type"].unique().tolist()
if "proportion_seropositiveXcluster_" in output
]
max_value = -10.0
if len(sero_outputs) == 0:
cluster_axis.text(
0.0, 0.5,
"Cluster-specific seroprevalence outputs are not available for this run"
)
else:
for output in sero_outputs:
output_mask = df["type"] == output
cluster = output.split("proportion_seropositiveXcluster_")[1]
seroprevalence_by_cluster[cluster] = {}
for q in quantile_vals:
q_mask = df["quantile"] == q
seroprevalence_by_cluster[cluster][q] = [
100.0 * v
for v in df[output_mask][q_mask]["value"].tolist()
]
q_keys = requested_quantiles if requested_quantiles else sorted(
quantile_vals)
num_quantiles = len(q_keys)
half_length = num_quantiles // 2
cluster_names = [
get_plot_text_dict(i.split("proportion_seropositiveXcluster_")[1])
for i in sero_outputs
]
lower_q_key = (100.0 - credible_range) / 100.0 / 2.0
upper_q_key = 1.0 - lower_q_key
x_positions = range(len(seroprevalence_by_cluster))
for i, cluster in enumerate(list(seroprevalence_by_cluster.keys())):
cluster_axis.plot(
[x_positions[i], x_positions[i]],
[
seroprevalence_by_cluster[cluster][lower_q_key],
seroprevalence_by_cluster[cluster][upper_q_key],
],
"-",
color="black",
lw=1.0,
)
cluster_axis.set_xticklabels(cluster_names,
fontsize=10,
rotation=90)
max_value = max(max_value,
seroprevalence_by_cluster[cluster][upper_q_key][0])
if num_quantiles % 2:
q_key = q_keys[half_length]
label = None if i > 0 else "model"
cluster_axis.plot(
x_positions[i],
seroprevalence_by_cluster[cluster][q_key],
"o",
color="black",
markersize=4,
label=label,
)
cluster_axis.xaxis.set_ticks(x_positions)
cluster_axis.set_ylim(bottom=0.0)
cluster_axis.set_ylabel("% previously infected", fontsize=13)
_date = ref_date + datetime.timedelta(days=time)
axis, max_value, df, seroprevalence_by_age = plot_age_seroprev_to_axis(
uncertainty_df,
scenario_id,
time,
age_axis,
requested_quantiles,
ref_date,
name,
add_date_as_title=False,
add_ylabel=False,
)
plotter.save_figure(fig,
filename="sero_by_cluster",
subdir="outputs",
title_text="")
overall_seropos_estimates = df[df["type"] == "proportion_seropositive"][[
"quantile", "value"
]].set_index("quantile")
return max_value, seroprevalence_by_cluster, overall_seropos_estimates
def plot_timeseries_with_uncertainty(plotter: Plotter,
uncertainty_df: pd.DataFrame,
output_name: str,
scenario_idxs: List[int],
targets: dict,
is_logscale=False,
x_low=0.0,
x_up=1e6,
axis=None,
n_xticks=None,
ref_date=REF_DATE,
add_targets=True,
overlay_uncertainty=True,
title_font_size=12,
label_font_size=10,
dpi_request=300,
capitalise_first_letter=False,
legend=False,
requested_x_ticks=None,
show_title=True,
ylab=None,
x_axis_to_date=True,
start_quantile=0,
sc_colors=None,
custom_title=None,
vlines={},
hlines={}):
"""
Plots the uncertainty timeseries for one or more scenarios.
Also plots any calibration targets that are provided.
"""
single_panel = axis is None
if single_panel:
fig, axis, _, _, _, _ = plotter.get_figure()
n_scenarios_to_plot = len(scenario_idxs)
if sc_colors is None:
n_scenarios_to_plot = min([len(scenario_idxs), len(COLORS)])
colors = _apply_transparency(COLORS[:n_scenarios_to_plot],
ALPHAS[:n_scenarios_to_plot])
# Plot each scenario on a single axis
data_to_return = {}
for i, scenario_idx in enumerate(scenario_idxs[:n_scenarios_to_plot]):
if sc_colors is None:
if scenario_idx < len(colors):
scenario_colors = colors[scenario_idx]
else:
scenario_colors = colors[-1]
else:
scenario_colors = sc_colors[i]
times, quantiles = _plot_uncertainty(
axis,
uncertainty_df,
output_name,
scenario_idx,
x_up,
x_low,
scenario_colors,
overlay_uncertainty=overlay_uncertainty,
start_quantile=start_quantile,
zorder=i + 1,
)
data_to_return[scenario_idx] = pd.DataFrame.from_dict(quantiles)
data_to_return[scenario_idx].insert(0, "days from 31/12/2019", times)
# Add plot targets
if add_targets:
values, times = _get_target_values(targets, output_name)
trunc_values = [
v for (v, t) in zip(values, times) if x_low <= t <= x_up
]
trunc_times = [
t for (v, t) in zip(values, times) if x_low <= t <= x_up
]
_plot_targets_to_axis(axis,
trunc_values,
trunc_times,
on_uncertainty_plot=True)
# Sort out x-axis
if x_axis_to_date:
change_xaxis_to_date(axis, ref_date, rotation=0)
axis.tick_params(axis="x", labelsize=label_font_size)
axis.tick_params(axis="y", labelsize=label_font_size)
# Add lines with marking text to plots
add_vertical_lines_to_plot(axis, vlines)
add_horizontal_lines_to_plot(axis, hlines)
if output_name == "proportion_seropositive":
axis.yaxis.set_major_formatter(mtick.PercentFormatter(1, symbol=""))
if show_title:
title = custom_title if custom_title else get_plot_text_dict(
output_name)
axis.set_title(title, fontsize=title_font_size)
if requested_x_ticks is not None:
pyplot.xticks(requested_x_ticks)
elif n_xticks is not None:
pyplot.locator_params(axis="x", nbins=n_xticks)
if is_logscale:
axis.set_yscale("log")
elif not (output_name.startswith("rel_diff")
or output_name.startswith("abs_diff")):
axis.set_ylim(ymin=0)
if ylab is not None:
axis.set_ylabel(ylab, fontsize=label_font_size)
if legend:
pyplot.legend(labels=scenario_idxs)
if single_panel:
idx_str = "-".join(map(str, scenario_idxs))
filename = f"uncertainty-{output_name}-{idx_str}"
plotter.save_figure(fig, filename=filename, dpi_request=dpi_request)
return data_to_return
def plot_param_vs_param(
plotter: Plotter,
mcmc_params: List[pd.DataFrame],
parameters: list,
burn_in: int,
style: str,
bins: int,
label_font_size: int,
label_chars: int,
dpi_request: int,
label_param_string=True,
show_ticks=False,
file_name="parameter_correlation_matrix",
):
"""
Plot the parameter correlation matrices for each parameter combination.
"""
# Prelims
fig, axes, _, _, _, _ = plotter.get_figure(n_panels=len(parameters)**2)
row_data, col_data = {}, {}
# Get x and y data separately and collate up over the chains
for row_idx, row_param_name in enumerate(parameters):
row_data[row_param_name] = []
for chain in range(len(mcmc_params)):
x_param_mask = (mcmc_params[chain]["name"] == row_param_name) & (
mcmc_params[chain]["run"] > burn_in)
row_data[row_param_name] += mcmc_params[chain][x_param_mask][
"value"].to_list()
for col_idx, col_param_name in enumerate(parameters):
col_data[col_param_name] = []
for chain in range(len(mcmc_params)):
y_param_mask = (mcmc_params[chain]["name"] == col_param_name) & (
mcmc_params[chain]["run"] > burn_in)
col_data[col_param_name] += mcmc_params[chain][y_param_mask][
"value"].to_list()
# Loop over parameter combinations
for row_idx, row_param_name in enumerate(parameters):
for col_idx, col_param_name in enumerate(parameters):
axis = axes[row_idx, col_idx]
if not show_ticks:
axis.xaxis.set_ticks([])
axis.yaxis.set_ticks([])
else:
axis.tick_params(labelsize=4)
# Plot
if row_idx > col_idx:
if style == "Scatter":
axis.scatter(
col_data[col_param_name],
row_data[row_param_name],
alpha=0.5,
s=0.1,
color="k",
)
elif style == "KDE":
sns.kdeplot(
col_data[col_param_name],
row_data[row_param_name],
ax=axis,
shade=True,
levels=5,
lw=1.0,
)
else:
axis.hist2d(col_data[col_param_name],
row_data[row_param_name],
bins=bins)
elif row_idx == col_idx:
axis.hist(
row_data[row_param_name],
color=[0.2, 0.2, 0.6] if style == "Shade" else "k",
bins=bins,
)
axis.yaxis.set_ticks([])
else:
axis.axis("off")
# Axis labels (these have to be reversed for some reason)
x_param_label = col_param_name if label_param_string else str(
col_idx + 1)
y_param_label = row_param_name if label_param_string else str(
row_idx + 1)
if row_idx == len(parameters) - 1:
axis.set_xlabel(get_plot_text_dict(x_param_label),
fontsize=label_font_size,
labelpad=3)
if col_idx == 0:
axis.set_ylabel(get_plot_text_dict(y_param_label),
fontsize=label_font_size)
# Save
plotter.save_figure(fig, filename=file_name, dpi_request=dpi_request)
