def show_best_runs_df(best_pr, filters, save_csv=True):
hd(
"h3",
f"Top 50 precisions at min_recall={filters.min_recall} {filters.classifier}",
)
if save_csv:
csv_filename = f"./report_best_pr__{munch_abbreviation_string(filters)}.csv"
best_pr.to_csv(csv_filename, index=False, float_format="%g")
print(f"(All {len(best_pr)} rows exported to {csv_filename})")
pd.set_option("display.max_columns", None)
display(best_pr.head(50))
def plot_flu_info(run, flu, peps_prs_df, min_recall=0.005, classifier=None):
"""
flu: string fluorosequence e.g. '..1......;0,0,0'
peps_prs_df: the DataFrame from a notebook containing precision,recall,score
for the peptides you're interested in.
classifier: None to use any available preferred classifier, or one of the
supported classifiers in RunResult::get_available_classifiers(),
e.g. 'rf', 'nn'
"""
hd("h2", f"flu_info for: {flu}")
cb = run.test_call_bag(classifier=classifier)
pf2 = cb.peps__pepstrs__flustrs__p2()
pf2 = pf2[pf2.flustr == flu]
pep_iz = pf2.pep_i.values
print(f"Peptides: {pep_iz}")
if peps_prs_df.empty:
# The peps_prs_df that is passed to us is typically the large
# DataFrame with PRS information for ALL peptides across ALL
# runs, but if this is a PTM notebook, it is often the case
# that PRS info is only loaded for peptides with PTM locations.
# This DataFrame has lots of other information, which is what
# we actually want to display, otherwise we'd just recompute
# the PRS for these peptides here.
print("No PR info found for peps. (No PTM peps with this flu?)")
else:
df_pr = peps_prec_at_min_recall_df(peps_prs_df, min_recall=min_recall)
display(df_pr)
# See the fn in plots_dev_mhc which calls the lower-level _plot_pr_curve using the
# prs info we already have in the df. The following call recomputes the PR all over
# again for these peps!
plots.plot_pr_breakout(
run,
pep_iz=pep_iz,
_size=500,
f_title=f"PR for peptides with flu {flu} ({cb.classifier_name})",
)
def print_titles(filters):
hd("h1", "Best runs per peptide")
hd("h3", "Filters")
json_print(filters)
print()