def _base_scaling(plt_df: pd.DataFrame,
sweep_vars: Optional[Sequence[str]] = None,
with_baseline: bool = True) -> gg.ggplot:
"""Base underlying piece of the scaling plots for deep sea."""
p = (gg.ggplot(plt_df) + gg.aes(x='size', y='episode'))
if np.all(plt_df.finished):
p += gg.geom_point(gg.aes(colour='solved'), size=3, alpha=0.75)
else:
p += gg.geom_point(gg.aes(shape='finished', colour='solved'),
size=3,
alpha=0.75)
p += gg.scale_shape_manual(values=['x', 'o'])
if np.all(plt_df.solved):
p += gg.scale_colour_manual(values=['#313695']) # blue
else:
p += gg.scale_colour_manual(values=['#d73027',
'#313695']) # [red, blue]
if with_baseline:
baseline_df = _make_baseline(plt_df, sweep_vars)
p += gg.geom_line(data=baseline_df,
colour='black',
linetype='dashed',
alpha=0.4,
size=1.5)
return p
def ensemble_plot(experiment_name='ensemble_nn', data_path=_DEFAULT_DATA_PATH):
"""Specialized plotting script for TS tutorial paper ensemble NN."""
df = load_data(experiment_name, data_path)
plt_df = (df.groupby(['agent', 't']).agg({
'instant_regret': np.mean
}).reset_index())
def _get_agent_family(agent_name):
if 'dropout' in agent_name.lower():
return 'Dropout'
elif 'ensemble' in agent_name.lower():
return 'Ensemble'
elif '/' in agent_name.lower():
return 'Annealing epsilon'
else:
return 'Fixed epsilon'
def _rename_ensemble(agent_name):
if 'ensemble' in agent_name:
n_ensemble = agent_name.split('-')[0]
new_name = 'ensemble=' + n_ensemble.zfill(3)
return new_name
else:
return agent_name
plt_df['agent_name'] = plt_df.agent.apply(_rename_ensemble)
plt_df['agent_family'] = plt_df.agent.apply(_get_agent_family)
custom_colors = ['#d53e4f', '#fdae61', '#a6d96a', '#66c2a5', '#5e4fa2']
plot_dict = {}
for agent_family, df_family in plt_df.groupby(['agent_family']):
if agent_family == 'Ensemble':
custom_labels = [
'Ensemble 3', 'Ensemble 10', 'Ensemble 30', 'Ensemble 100',
'Ensemble 300'
]
gg_legend = gg.scale_colour_manual(values=custom_colors,
labels=custom_labels,
name='Agent')
else:
gg_legend = gg.scale_colour_manual(custom_colors, name='Agent')
p = (gg.ggplot(df_family) +
gg.aes('t', 'instant_regret', colour='agent_name') +
gg.geom_line(size=1.25, alpha=0.75) +
gg.facet_wrap('~ agent_family') + gg_legend +
gg.coord_cartesian(ylim=(0, 60)) + gg.xlab('Timestep (t)') +
gg.ylab('Average instantaneous regret') +
gg.theme(figure_size=(6, 6)))
plot_dict[experiment_name + '_' + agent_family] = p
return plot_dict
def bsuite_bar_plot(df_in: pd.DataFrame,
sweep_vars: Sequence[str] = None) -> gg.ggplot:
"""Output bar plot of bsuite data."""
df = _clean_bar_plot_data(df_in, sweep_vars)
p = (gg.ggplot(df)
+ gg.aes(x='env', y='score', colour='type', fill='type')
+ gg.geom_bar(position='dodge', stat='identity')
+ gg.geom_hline(yintercept=1., linetype='dashed', alpha=0.5)
+ gg.scale_colour_manual(plotting.CATEGORICAL_COLOURS)
+ gg.scale_fill_manual(plotting.CATEGORICAL_COLOURS)
+ gg.xlab('experiment')
+ gg.theme(axis_text_x=gg.element_text(angle=25, hjust=1))
)
if not all(df.finished): # add a layer of alpha for unfinished jobs
p += gg.aes(alpha='finished')
p += gg.scale_alpha_discrete(range=[0.3, 1.0])
# Compute the necessary size of the plot
if sweep_vars:
p += gg.facet_wrap(sweep_vars, labeller='label_both', ncol=1)
n_hypers = df[sweep_vars].drop_duplicates().shape[0]
else:
n_hypers = 1
return p + gg.theme(figure_size=(14, 3 * n_hypers + 1))
def plot_individual_returns(
df_in: pd.DataFrame,
max_episode: int,
return_column: str = 'episode_return',
colour_var: Optional[str] = None,
yintercept: Optional[float] = None,
sweep_vars: Optional[Sequence[str]] = None) -> gg.ggplot:
"""Plot individual learning curves: one curve per sweep setting."""
df = df_in.copy()
df['unique_group'] = _make_unique_group_col(df, sweep_vars)
p = (gg.ggplot(df) +
gg.aes(x='episode', y=return_column, group='unique_group') +
gg.coord_cartesian(xlim=(0, max_episode)))
if colour_var:
p += gg.geom_line(gg.aes(colour=colour_var), size=1.1, alpha=0.75)
if len(df[colour_var].unique()) <= 5:
df[colour_var] = df[colour_var].astype('category')
p += gg.scale_colour_manual(values=FIVE_COLOURS)
else:
p += gg.geom_line(size=1.1, alpha=0.75, colour='#313695')
if yintercept:
p += gg.geom_hline(yintercept=yintercept,
alpha=0.5,
size=2,
linetype='dashed')
return facet_sweep_plot(p, sweep_vars, tall_plot=True)
def plot_time_curve_with_threshold(self):
toplot = self.aggregated.melt(
id_vars='hour',
value_vars=['number_bacteria', 'number_actin'],
value_name='counts',
var_name='Object')
colors = self.create_color_list()
myfig = (
ggplot(toplot, aes("hour", "counts", color="Object")) +
geom_point() + geom_line() + labels.xlab("Time [hours]") +
labels.ylab("Average number of objects/nuclei") +
pn.scale_colour_manual(values=colors,
labels=list(self.sel_channel_time.value),
name="") + pn.labs(colour="") +
pn.scale_x_continuous(
breaks=np.sort(self.result.hour.unique()),
labels=list(np.sort(self.result.hour.unique()).astype(str))))
self.time_curve_fig = myfig
self.out_plot2.clear_output()
with self.out_plot2:
display(myfig)
def plot_time_curve_by_channel(self, b=None):
"""Callback to polot time curve of number of bacteria/nuclei for
each selected channel. Called by plot_time_curve_button."""
if self.aggregated is None:
self.data_aggregation()
if len(self.sel_channel_time.value) == 0:
print("Select at least one channel")
else:
subset = self.aggregated[self.aggregated.channel.isin(
self.sel_channel_time.value)].copy(deep=True)
subset.loc[:, "channel"] = subset.channel.astype(
pd.CategoricalDtype(self.sel_channel_time.value, ordered=True))
colors = self.create_color_list()
myfig = (
ggplot(subset, aes("hour", "normalized", color="channel")) +
geom_point() + geom_line() + labels.xlab("Time [hours]") +
labels.ylab("Average number of bacteria/nuclei") +
pn.scale_colour_manual(
values=colors,
labels=list(self.sel_channel_time.value),
name="") + pn.labs(colour="") + pn.scale_x_continuous(
breaks=np.sort(self.result.hour.unique()),
labels=list(
np.sort(self.result.hour.unique()).astype(str))))
self.time_curve_fig = myfig
self.out_plot2.clear_output()
with self.out_plot2:
display(myfig)
def __plot(
self,
plot_data,
x,
y,
colour,
lbl_x,
lbl_y,
facet,
facet_scales,
facet_by,
smoothed,
points,
error_bars,
save,
):
cbbPalette = [
"#000000",
"#E69F00",
"#56B4E9",
"#009E73",
"#0072B2",
"#D55E00",
"#CC79A7",
]
plt = ggplot(data=plot_data, mapping=aes(x=x, y=y, colour=colour))
plt += xlab(lbl_x)
plt += ylab(lbl_y)
# + facet_grid("site~", scales="free")
# + geom_line()
if facet:
# TODO: use facet as save
nrow, ncol = self.get_facet_rows(plot_data, facet_by)
plt += facet_wrap(facet_by, nrow=nrow, ncol=ncol, scales=facet_scales)
if points:
plt += geom_point()
if error_bars:
# TODO use generic way to compute them
pass
# self.plt += geom_errorbar(aes(ymin="ACI_mean - ACI_std", ymax="ACI_mean + ACI_std"))
# TODO: use smooth as save
if smoothed:
plt += geom_smooth(
method="mavg",
se=False,
method_args={"window": 4, "center": True, "min_periods": 1},
)
else:
plt += geom_line()
plt += scale_colour_manual(values=cbbPalette, guide=False)
plt += scale_x_continuous(labels=label_x)
plt += theme(figure_size=(15, 18), dpi=150)
if save:
plt.save(**save)
return plt
def _plot_regret_group(df: pd.DataFrame, group_col: str) -> gg.ggplot:
"""Plots the average regret through time when grouped."""
group_name = group_col.replace('_', ' ')
df[group_name] = df[group_col].astype('category')
p = (gg.ggplot(df)
+ gg.aes(x='episode', y='average_regret',
group=group_name, colour=group_name, fill=group_name)
+ gg.geom_smooth(method=smoothers.mean, span=0.1, size=1.75, alpha=0.1)
+ gg.scale_colour_manual(values=FIVE_COLOURS)
+ gg.scale_fill_manual(values=FIVE_COLOURS))
return p
def _bar_plot_compare(df: pd.DataFrame) -> gg.ggplot:
"""Bar plot of buite score data, comparing agents on each experiment."""
p = (gg.ggplot(df) +
gg.aes(x='agent', y='score', colour='agent', fill='agent') +
gg.geom_bar(position='dodge', stat='identity') +
gg.geom_hline(yintercept=1., linetype='dashed', alpha=0.5) +
gg.theme(axis_text_x=gg.element_text(angle=25, hjust=1)) +
gg.scale_colour_manual(plotting.CATEGORICAL_COLOURS) +
gg.scale_fill_manual(plotting.CATEGORICAL_COLOURS))
if not all(df.finished): # add a layer of alpha for unfinished jobs
p += gg.aes(alpha='finished')
p += gg.scale_alpha_discrete(range=[0.3, 1.0])
return p
def plot_regret(df_in: pd.DataFrame,
sweep_vars: Sequence[Text] = None) -> gg.ggplot:
"""Plot average regret of deep_sea through time by size."""
df = df_in.copy()
df = df[df['size'].isin([10, 20, 30, 40, 50])]
df['avg_bad'] = df.total_bad_episodes / df.episode
df['size'] = df['size'].astype('category')
p = (
gg.ggplot(df[df.episode <= NUM_EPISODES]) +
gg.aes('episode', 'avg_bad', group='size', colour='size') +
gg.geom_line(size=2, alpha=0.75) + gg.geom_hline(
gg.aes(yintercept=0.99), linetype='dashed', alpha=0.4, size=1.75) +
gg.geom_hline(gg.aes(yintercept=0.0), alpha=0) # axis hack
+ gg.ylab('average bad episodes') +
gg.scale_colour_manual(values=plotting.FIVE_COLOURS))
return plotting.facet_sweep_plot(p, sweep_vars)
def plot_regret_ave_scaling(df_in: pd.DataFrame,
group_col: str,
episode: int,
regret_thresh: float,
sweep_vars: Sequence[str] = None,
regret_col: str = 'total_regret') -> gg.ggplot:
"""Point plot of average regret investigating scaling to threshold."""
df = _preprocess_ave_regret(df_in, group_col, episode, sweep_vars, regret_col)
group_name = group_col.replace('_', ' ')
p = (gg.ggplot(df)
+ gg.aes(x=group_name, y='average_regret',
colour='average_regret < {}'.format(regret_thresh))
+ gg.geom_point(size=5, alpha=0.8)
+ gg.scale_x_log10(breaks=[1, 3, 10, 30, 100])
+ gg.scale_colour_manual(values=['#d73027', '#313695'])
+ gg.ylab('average regret at {} episodes'.format(episode))
+ gg.geom_hline(gg.aes(yintercept=0.0), alpha=0) # axis hack
)
return facet_sweep_plot(p, sweep_vars)
def plot_scale(df: pd.DataFrame,
sweep_vars: Sequence[str] = None) -> gg.ggplot:
"""Plots the best episode observed by height_threshold."""
df = cp_swingup_preprocess(df_in=df)
group_vars = ['height_threshold']
if sweep_vars:
group_vars += sweep_vars
plt_df = df.groupby(group_vars)['best_episode'].max().reset_index()
p = (
gg.ggplot(plt_df) +
gg.aes(x='factor(height_threshold)',
y='best_episode',
colour='best_episode > {}'.format(GOOD_EPISODE)) +
gg.geom_point(size=5, alpha=0.8) +
gg.scale_colour_manual(values=['#d73027', '#313695']) +
gg.geom_hline(gg.aes(yintercept=0.0), alpha=0) # axis hack
+ gg.scale_x_discrete(breaks=[0, 0.25, 0.5, 0.75, 1.0]) +
gg.ylab('best return in first {} episodes'.format(NUM_EPISODES)) +
gg.xlab('height threshold'))
return plotting.facet_sweep_plot(p, sweep_vars)
def label_x(dates):
res = [(datetime.datetime(2018, 1, 1) + datetime.timedelta(x)).strftime("%d-%m") for x in dates]
print(res)
return res
(ggplot(data=res, mapping=aes(x='julian', y='value', colour='type'))
+ xlab("Day")
+ ylab("Mean number of detected songs")
+ facet_grid("type~", scales="free")
# + geom_line()
# + facet_wrap("type", nrow=2, ncol=1)
+ geom_point()
# + geom_errorbar(aes(ymin="ACI_mean - ACI_std", ymax="ACI_mean + ACI_std"))
+ geom_smooth(method="mavg", se=False, method_args={"window": 4, "center": True, "min_periods": 1})
+ scale_colour_manual(values=cbbPalette, guide=False)
+ scale_x_continuous(labels=label_x)).save("figs/song_events_aci_BARROW_mean_smoothed.png", height=10, width=16, dpi=150)
(ggplot(data=res, mapping=aes(x='julian', y='n_events_sum', colour='site'))
+ xlab("Day")
+ ylab("Total number of detected songs")
# + facet_grid("site~", scales="free")
# + facet_wrap("site", nrow=2, ncol=3)
+ geom_point()
# + geom_errorbar(aes(ymin="ACI_mean - ACI_std", ymax="ACI_mean + ACI_std"))
+ geom_smooth(method="mavg", se=False, method_args={"window": 4, "center": True, "min_periods": 1})
+ scale_colour_manual(values=cbbPalette, guide=False)
+ scale_x_continuous(labels=label_x)).save("figs/song_events_BARW0_sum.png", height=10, width=16, dpi=150)
#################
### Denoising ###
na_rm=False)
else:
g += p9.geom_crossbar(p9.aes(x="x",
y='center',
ymin='low',
ymax='high',
group="factor(group_x)",
colour="factor(group)",
fill="factor(group)"),
position=p9.position_dodge(
0.7, preserve='single'),
na_rm=True,
alpha=0.2)
g += p9.scale_fill_manual(values=ez_colors(g.n_groups('group')))
g += p9.scale_colour_manual(values=ez_colors(g.n_groups('group')))
elif geom == 'ribbon':
g = EZPlot(gdata.dropna())
# set groups
if group is None:
g += p9.geom_ribbon(p9.aes(x="x",
y='center',
ymin='low',
ymax='high'),
fill=ez_colors(1)[0],
alpha=0.2,
na_rm=False)
g += p9.geom_line(p9.aes(x="x", y='center'),
input_data_UMAPencoded_df['dataset'] = 'original'
simulated_data_UMAPencoded_df['dataset'] = 'simulated'
# Concatenate input and simulated dataframes together
combined_data_df = pd.concat(
[input_data_UMAPencoded_df, simulated_data_UMAPencoded_df])
# Plot
g_input_sim = ggplot(combined_data_df, aes(x='1', y='2')) + geom_point(aes(color='dataset'), alpha=0.3) + labs(x = "UMAP 1", y = "UMAP 2", title = "UMAP of original and simulated data") + theme_bw() + theme(
legend_title_align = "center",
plot_background=element_rect(fill='white'),
legend_key=element_rect(fill='white', colour='white'),
plot_title=element_text(weight='bold')
) \
+ guides(colour=guide_legend(override_aes={'alpha': 1})) \
+ scale_colour_manual(["grey", '#87CEFA'])
print(g_input_sim)
ggsave(plot=g_input_sim, filename=umap_overlay_file, dpi=300)
# ## 2. Visualize effects of multiple experiments in PCA space
# In[13]:
get_ipython().run_cell_magic(
'time', '',
'\nall_data_df = pd.DataFrame()\n\n# Get batch 1 data\npartition_1_file = os.path.join(\n partition_dir,\n "Partition_1.txt.xz")\n\npartition_1 = pd.read_table(\n partition_1_file,\n header=0,\n index_col=0,\n sep=\'\\t\')\n\n\nfor i in lst_num_partitions:\n print(\'Plotting PCA of 1 partition vs {} partition...\'.format(i))\n \n # Simulated data with all samples in a single partition\n original_data_df = partition_1.copy()\n \n # Add grouping column for plotting\n original_data_df[\'num_partitions\'] = \'1\'\n \n # Get data with additional partitions added\n partition_other_file = os.path.join(\n partition_dir,\n "Partition_"+str(i)+".txt.xz")\n\n partition_other = pd.read_table(\n partition_other_file,\n header=0,\n index_col=0,\n sep=\'\\t\')\n \n # Simulated data with i partitions\n partition_data_df = partition_other\n \n # Add grouping column for plotting\n partition_data_df[\'num_partitions\'] = \'multiple\'\n \n # Concatenate datasets together\n combined_data_df = pd.concat([original_data_df, partition_data_df])\n\n # PCA projection\n pca = PCA(n_components=2)\n\n # Encode expression data into 2D PCA space\n combined_data_numeric_df = combined_data_df.drop([\'num_partitions\'], axis=1)\n combined_data_PCAencoded = pca.fit_transform(combined_data_numeric_df)\n\n\n combined_data_PCAencoded_df = pd.DataFrame(combined_data_PCAencoded,\n index=combined_data_df.index,\n columns=[\'PC1\', \'PC2\']\n )\n \n # Variance explained\n print(pca.explained_variance_ratio_) \n \n # Add back in batch labels (i.e. labels = "batch_"<how many batch effects were added>)\n combined_data_PCAencoded_df[\'num_partitions\'] = combined_data_df[\'num_partitions\']\n \n # Add column that designates which batch effect comparision (i.e. comparison of 1 batch vs 5 batches\n # is represented by label = 5)\n combined_data_PCAencoded_df[\'comparison\'] = str(i)\n \n # Concatenate ALL comparisons\n all_data_df = pd.concat([all_data_df, combined_data_PCAencoded_df])\n \n # Plot individual comparisons\n print(ggplot(combined_data_PCAencoded_df, aes(x=\'PC1\', y=\'PC2\')) \\\n + geom_point(aes(color=\'num_partitions\'), alpha=0.2) \\\n + labs(x = "PC 1", y = "PC 2", title = "Partition 1 and Partition {}".format(i))\\\n + theme_bw() \\\n + theme(\n legend_title_align = "center",\n plot_background=element_rect(fill=\'white\'),\n legend_key=element_rect(fill=\'white\', colour=\'white\'), \n plot_title=element_text(weight=\'bold\')\n ) \\\n + guides(colour=guide_legend(override_aes={\'alpha\': 1})) \\\n + scale_colour_manual(["grey", \'#b3e5fc\'])\n ) '
)
# In[14]:
#select_data = combined_data_df[combined_data_df['experiment_id'] != 'Not selected']
# Plot
ggplot(combined_data_df, aes(x='1', y='2')) + geom_point(
aes(color='experiment_id'),
alpha=0.3) + facet_wrap('~dataset') + xlab('UMAP 1') + ylab(
'UMAP 2') + ggtitle('UMAP of original and simulated data (gene space)')
#+ xlim(3,12) \
#+ ylim(-7,10) \
#+ scale_colour_manual(values=["blue", "purple", "orange", "red", "magenta", "lightgrey"]) \
# In[12]:
# Overlay original and simulated data
ggplot(combined_data_df, aes(x='1', y='2')) + geom_point(
aes(color='dataset'), alpha=0.3) + scale_colour_manual(
values=["grey", "blue"]) + xlab('UMAP 1') + ylab('UMAP 2') + ggtitle(
'UMAP of original and simulated data (gene space)')
# ## Visualize simulated data (gene space) projected into PCA space
# In[13]:
# UMAP embedding of original input data
# Get and save model
pca = PCA(n_components=2)
pca.fit(normalized_data)
input_data_PCAencoded = pca.transform(normalized_data)
input_data_PCAencoded_df = pd.DataFrame(data=input_data_PCAencoded,
index=normalized_data.index,
