def get_order(answers, metrics, min_answers=50):
sessions = answers.groupby('session').apply(len)
sessions = sessions[sessions >= min_answers]
metric = answers["correct"]
ends = defaultdict(lambda: [])
for session in sessions.index:
answer_index = answers[answers['session'] == session].index
for name, metric in metrics.items():
a = metric.loc[answer_index]
success = rolling_success(a)
end = success[-1]
ends[name].append(end)
return pd.DataFrame.from_dict(ends)
def master_curves(answers, metrics, min_answers=50, student_count=None, smooth=0):
sessions = answers.groupby('session').apply(len)
sessions = sessions[sessions >= min_answers]
# sessions = sessions.sample(1)
for i, mcs in enumerate(metrics):
plt.subplot(len(metrics), 1, i + 1)
for metric_name, metric in mcs.items():
success_mean = metric.mean()
s = np.zeros(min_answers)
for student in sessions.index[:student_count] if student_count else sessions.index:
current = success_mean
a = metric.loc[answers[answers['session'] == student].index]
success = rolling_success(a, success_mean)
if smooth:
success = pd.rolling_mean(pd.Series(success), 20) # hack to make line more smooth
s += success[:min_answers]
plt.plot(range(len(s)), s / (student_count if student_count else len(sessions)), label=metric_name)
plt.legend()