def plot_1_1(ax: matplotlib.axes.SubplotBase, df_pred_str: pd.DataFrame, df_pred_str_null: pd.DataFrame, rows_set: np.ndarray) -> None:
mu, sigma = ClusteringUtils.calc_distribution(df_pred_str.loc[rows_set])
mu_null, sigma_null = ClusteringUtils.calc_distribution(df_pred_str_null.loc[rows_set])
y = pd.DataFrame([
mu - sigma,
mu,
mu + sigma,
mu_null - sigma_null,
mu_null,
mu_null + sigma_null
]).transpose()
x = pd.DataFrame([df_pred_str.columns for _ in range(len(y.columns))]).transpose()
lty = ["--", "-", "--", "--", "-", "--"]
lwd = [1, 2, 1, 1, 2, 1]
colors = ["blue", "blue", "blue", "red", "red", "red"]
legend_lines = [
Line2D([0], [0], color="blue", linestyle="-", linewidth=2),
Line2D([0], [0], color="red", linestyle="-", linewidth=2),
Line2D([0], [0], color="gray", linestyle="--", linewidth=1)
]
for col in x.columns:
ax.plot(x[col], y[col], color=colors[col], linestyle=lty[col], linewidth=lwd[col])
ax.set_xlabel("k")
ax.set_ylabel("prediction strength")
ax.legend(legend_lines, ["model", "null", "95% CI"], loc='upper right')
def stackplot(
self,
eco,
title: titleType = None,
logscale: bool = True,
ax: matplotlib.axes.SubplotBase = None,
) -> matplotlib.figure.Figure:
populations = eco.population_sizes
if ax is None:
_, ax = plt.subplots()
else:
ax = ax
figure = ax.get_figure()
turns = range(len(populations))
pops = [
[populations[iturn][ir] for iturn in turns]
for ir in self.result_set.ranking
]
ax.stackplot(turns, *pops)
ax.yaxis.tick_left()
ax.yaxis.set_label_position("right")
ax.yaxis.labelpad = 25.0
ax.set_ylim([0.0, 1.0])
ax.set_ylabel("Relative population size")
ax.set_xlabel("Turn")
if title is not None:
ax.set_title(title)
trans = transforms.blended_transform_factory(ax.transAxes, ax.transData)
ticks = []
for i, n in enumerate(self.result_set.ranked_names):
x = -0.01
y = (i + 0.5) * 1 / self.result_set.num_players
ax.annotate(
n,
xy=(x, y),
xycoords=trans,
clip_on=False,
va="center",
ha="right",
fontsize=5,
)
ticks.append(y)
ax.set_yticks(ticks)
ax.tick_params(direction="out")
ax.set_yticklabels([])
if logscale:
ax.set_xscale("log")
plt.tight_layout()
return figure
def stackplot(
self,
eco,
title: titleType = None,
logscale: bool = True,
ax: matplotlib.axes.SubplotBase = None,
) -> matplotlib.figure.Figure:
populations = eco.population_sizes
if ax is None:
_, ax = plt.subplots()
else:
ax = ax
figure = ax.get_figure()
turns = range(len(populations))
pops = [
[populations[iturn][ir] for iturn in turns]
for ir in self.result_set.ranking
]
ax.stackplot(turns, *pops)
ax.yaxis.tick_left()
ax.yaxis.set_label_position("right")
ax.yaxis.labelpad = 25.0
ax.set_ylim([0.0, 1.0])
ax.set_ylabel("Relative population size")
ax.set_xlabel("Turn")
if title is not None:
ax.set_title(title)
trans = transforms.blended_transform_factory(ax.transAxes, ax.transData)
ticks = []
for i, n in enumerate(self.result_set.ranked_names):
x = -0.01
y = (i + 0.5) * 1 / self.result_set.num_players
ax.annotate(
n,
xy=(x, y),
xycoords=trans,
clip_on=False,
va="center",
ha="right",
fontsize=5,
)
ticks.append(y)
ax.set_yticks(ticks)
ax.tick_params(direction="out")
ax.set_yticklabels([])
if logscale:
ax.set_xscale("log")
plt.tight_layout()
return figure
def plot_1_2(ax: matplotlib.axes.SubplotBase, diff_df: pd.DataFrame, row_id: int) -> None:
ax.bar(diff_df.columns, diff_df.loc[row_id], color="red")
ax.set_title("difference")
ax.set_xlabel("k")
ax.set_ylabel("diff")
diff_df_min = diff_df.min().min()
diff_df_max = diff_df.max().max()
if CVResultsAggregator.check_limit(diff_df_min) and CVResultsAggregator.check_limit(diff_df_max):
ax.set_ylim(diff_df_min - 1, diff_df_max + 1)
def _plot_roc(self, axis: mpl.axes.SubplotBase, y_true: np.ndarray,
y_pred: np.ndarray, label: str, color: str):
x, y, _ = roc_curve(y_true, y_pred)
axis.plot(x,
y,
color,
label="{label}, area={auc:.2f}".format(auc=auc(x, y),
label=label))
axis.plot([0, 1], [0, 1], 'k--')
axis.set_xlabel("False Positive Rate")
axis.set_ylabel("True Positive Rate")
axis.set_title("ROC curves - {label}".format(label=label))
def plot_2_2(ax: matplotlib.axes.SubplotBase, df_eff_k: pd.DataFrame, rows_set: np.ndarray) -> None:
lty = ["--", "-", "--"]
lwd = [1, 2, 1]
mu, sigma = ClusteringUtils.calc_distribution(df_eff_k.loc[rows_set])
y = pd.DataFrame([mu - sigma, mu, mu + sigma]).transpose()
for col in y.columns:
ax.plot(df_eff_k.columns, y[col], color="red", linestyle=lty[col], linewidth=lwd[col])
ax.set_xlabel("k")
ax.set_ylabel("effective k")
ax.set_xlim(1, df_eff_k.columns.max())
ax.set_ylim(1, df_eff_k.columns.max())
abline_vals = np.array(ax.get_xlim())
ax.plot(abline_vals, abline_vals, color="grey", linestyle="--")
def plot_2_1_ci(ax: matplotlib.axes.SubplotBase, df_pred_str: pd.DataFrame, df_pred_str_null: pd.DataFrame,
rows_set: np.ndarray) -> None:
mu, sigma = ClusteringUtils.calc_distribution(df_pred_str.loc[rows_set])
mu_null, sigma_null = ClusteringUtils.calc_distribution(df_pred_str_null.loc[rows_set])
y = pd.DataFrame([
mu - sigma > mu_null + sigma_null
])
ax.imshow(y)
ax.set_xlabel("k")
ax.set_ylabel("CI intersection")
ax.set_yticks([])
ax.xaxis.set_major_locator(ticker.FixedLocator((np.arange(len(y.columns)))))
ax.xaxis.set_major_formatter(ticker.FixedFormatter(y.columns))
def plt_settings_axes(g: matplotlib.axes.SubplotBase,
count_df: dask.dataframe.core.DataFrame,
grouping_col: list, facet: str, hide_xtitle: bool,
log_y: bool) -> None:
"""
Helper function for plot settings, used in function plt_generic_1d.
Modifies parameter g for setting titles, axis, formats, etc.
:param g: matplotlib Axes which will be modified directly in the function.
:param count_df: pandas dataframe which is plotted.
:param grouping_col: column for x axis.
:param facet: parameter passed by function plt_generic_1d, giving information
on whether we are plotting and average or a count value (on y axis).
:param hide_xtitle: if set to True, doesn't display title for x axis
:param log_y: if set to True, plot in logarithmic scale (for y axis)
:return: nothing. changes are done directly by modifying parameter g.
"""
if facet not in ['freq', 'avg']:
raise ValueError(
'Parameter facet should be a string of value either "freq" or "avg"'
)
# SET X AXIS
# Labels
# no particular setup if number of labels is less than the first threshold
num_xlabels = len(count_df[grouping_col])
if num_xlabels < LABEL_THRESHOLD_ROTATION:
g.set_xticklabels(count_df[grouping_col])
# rotate by 90 degrees if number of labels is between first and second threshold
elif num_xlabels < LABEL_THRESHOLD_SELECT:
g.set_xticklabels(count_df[grouping_col], rotation=90)
# display only certain labels (and rotate by 45 degrees) if number of labels is higher
else:
number_of_steps = num_xlabels / 50
l = np.arange(0, num_xlabels, number_of_steps)
pos = (l / num_xlabels) * (max(g.get_xticks()) - min(g.get_xticks()))
g.set_xticks(pos)
g.set_xticklabels(count_df[grouping_col].iloc[l], rotation=45)
# Title
# option to remove the x axis title (when its obvious, e.g. for the years)
if hide_xtitle:
g.set_xlabel('')
else:
g.set_xlabel(grouping_col)
# SET Y AXIS
# log scale option
if log_y:
g.set_yscale("log")
if facet == 'freq':
g.set_ylabel('# content items (log scale)')
elif facet == 'avg':
g.set_ylabel('title length (log scale)')
else:
if facet == 'freq':
g.set_ylabel('# content items')
elif facet == 'avg':
g.set_ylabel('title length')
# Labels
ylabels = ['{:,.0f}'.format(y) for y in g.get_yticks()]
g.set_yticklabels(ylabels)
# Plot Title
if facet == 'freq':
g.set_title('Number of content items by %s' % grouping_col)
elif facet == 'avg':
g.set_title('Average title length of content items by %s' %
grouping_col)
def plot_2_1(ax: matplotlib.axes.SubplotBase, pval_df: pd.DataFrame, row_id: int) -> None:
ax.bar(pval_df.columns, -np.log10(pval_df.loc[row_id]), color="red")
ax.set_title("significance of difference")
ax.set_xlabel("k")
ax.set_ylabel("-log10(p-value)")
ax.set_ylim(0, 4)
