From 66a3c5b4861f99c965367be97816cc15f7bea24a Mon Sep 17 00:00:00 2001 From: Taha Yassine Kraiem Date: Wed, 23 Nov 2022 20:42:50 +0100 Subject: [PATCH] feat(chalice): funnel optimizations --- api/chalicelib/core/significance.py | 113 +++++++++++++--------------- 1 file changed, 53 insertions(+), 60 deletions(-) diff --git a/api/chalicelib/core/significance.py b/api/chalicelib/core/significance.py index 2abd87cf7..1522dc94a 100644 --- a/api/chalicelib/core/significance.py +++ b/api/chalicelib/core/significance.py @@ -181,9 +181,7 @@ def get_stages_and_events(filter_d, project_id) -> List[RealDictRow]: values=s["value"], value_key=f"value{i + 1}") n_stages_query.append(f""" (SELECT main.session_id, - {"MIN(main.timestamp)" if i + 1 < len(stages) else "MAX(main.timestamp)"} AS stage{i + 1}_timestamp, - '{event_type}' AS type, - '{s["operator"]}' AS operator + {"MIN(main.timestamp)" if i + 1 < len(stages) else "MAX(main.timestamp)"} AS stage{i + 1}_timestamp FROM {next_table} AS main {" ".join(extra_from)} WHERE main.timestamp >= {f"T{i}.stage{i}_timestamp" if i > 0 else "%(startTimestamp)s"} {f"AND main.session_id=T1.session_id" if i > 0 else ""} @@ -199,7 +197,8 @@ def get_stages_and_events(filter_d, project_id) -> List[RealDictRow]: n_stages_query += ") AS stages_t" n_stages_query = f""" - SELECT stages_and_issues_t.*, sessions.user_uuid FROM ( + SELECT stages_and_issues_t.*, sessions.user_uuid + FROM ( SELECT * FROM ( SELECT * FROM {n_stages_query} @@ -297,7 +296,21 @@ def pearson_corr(x: list, y: list): return r, confidence, False -def get_transitions_and_issues_of_each_type(rows: List[RealDictRow], all_issues_with_context, first_stage, last_stage): +# def tuple_or(t: tuple): +# x = 0 +# for el in t: +# x |= el # | is for bitwise OR +# return x +# +# The following function is correct optimization of the previous function because t is a list of 0,1 +def tuple_or(t: tuple): + for el in t: + if el > 0: + return 1 + return 0 + + +def get_transitions_and_issues_of_each_type(rows: List[RealDictRow], all_issues, first_stage, last_stage): """ Returns two lists with binary values 0/1: @@ -316,12 +329,6 @@ def get_transitions_and_issues_of_each_type(rows: List[RealDictRow], all_issues_ transitions = [] n_sess_affected = 0 errors = {} - for issue in all_issues_with_context: - split = issue.split('__^__') - errors[issue] = { - "errors": [], - "issue_type": split[0], - "context": split[1]} for row in rows: t = 0 @@ -329,38 +336,28 @@ def get_transitions_and_issues_of_each_type(rows: List[RealDictRow], all_issues_ last_ts = row[f'stage{last_stage}_timestamp'] if first_ts is None: continue - elif first_ts is not None and last_ts is not None: + elif last_ts is not None: t = 1 transitions.append(t) ic_present = False - for issue_type_with_context in errors: + for error_id in all_issues: + if error_id not in errors: + errors[error_id] = [] ic = 0 - issue_type = errors[issue_type_with_context]["issue_type"] - context = errors[issue_type_with_context]["context"] + issue_type = all_issues[error_id]["issue_type"] + context = all_issues[error_id]["context"] if row['issue_type'] is not None: if last_ts is None or (first_ts < row['issue_timestamp'] < last_ts): context_in_row = row['issue_context'] if row['issue_context'] is not None else '' if issue_type == row['issue_type'] and context == context_in_row: ic = 1 ic_present = True - errors[issue_type_with_context]["errors"].append(ic) + errors[error_id].append(ic) if ic_present and t: n_sess_affected += 1 - # def tuple_or(t: tuple): - # x = 0 - # for el in t: - # x |= el - # return x - def tuple_or(t: tuple): - for el in t: - if el > 0: - return 1 - return 0 - - errors = {key: errors[key]["errors"] for key in errors} all_errors = [tuple_or(t) for t in zip(*errors.values())] return transitions, errors, all_errors, n_sess_affected @@ -376,10 +373,9 @@ def get_affected_users_for_all_issues(rows, first_stage, last_stage): """ affected_users = defaultdict(lambda: set()) affected_sessions = defaultdict(lambda: set()) - contexts = defaultdict(lambda: None) + all_issues = {} n_affected_users_dict = defaultdict(lambda: None) n_affected_sessions_dict = defaultdict(lambda: None) - all_issues_with_context = set() n_issues_dict = defaultdict(lambda: 0) issues_by_session = defaultdict(lambda: 0) @@ -396,14 +392,13 @@ def get_affected_users_for_all_issues(rows, first_stage, last_stage): if iss is not None and (row[f'stage{last_stage}_timestamp'] is None or (row[f'stage{first_stage}_timestamp'] < iss_ts < row[f'stage{last_stage}_timestamp'])): context_string = row['issue_context'] if row['issue_context'] is not None else '' - issue_with_context = iss + '__^__' + context_string - contexts[issue_with_context] = {"context": context_string, "id": row["issue_id"]} - all_issues_with_context.add(issue_with_context) - n_issues_dict[issue_with_context] += 1 + if row["issue_id"] not in all_issues: + all_issues[row["issue_id"]] = {"context": context_string, "issue_type": row["issue_type"]} + n_issues_dict[row["issue_id"]] += 1 if row['user_uuid'] is not None: - affected_users[issue_with_context].add(row['user_uuid']) + affected_users[row["issue_id"]].add(row['user_uuid']) - affected_sessions[issue_with_context].add(row['session_id']) + affected_sessions[row["issue_id"]].add(row['session_id']) issues_by_session[row[f'session_id']] += 1 if len(affected_users) > 0: @@ -414,29 +409,28 @@ def get_affected_users_for_all_issues(rows, first_stage, last_stage): n_affected_sessions_dict.update({ iss: len(affected_sessions[iss]) for iss in affected_sessions }) - return all_issues_with_context, n_issues_dict, n_affected_users_dict, n_affected_sessions_dict, contexts + return all_issues, n_issues_dict, n_affected_users_dict, n_affected_sessions_dict def count_sessions(rows, n_stages): session_counts = {i: set() for i in range(1, n_stages + 1)} - for ind, row in enumerate(rows): + for row in rows: for i in range(1, n_stages + 1): if row[f"stage{i}_timestamp"] is not None: session_counts[i].add(row[f"session_id"]) + session_counts = {i: len(session_counts[i]) for i in session_counts} return session_counts def count_users(rows, n_stages): - users_in_stages = defaultdict(lambda: set()) - - for ind, row in enumerate(rows): + users_in_stages = {i: set() for i in range(1, n_stages + 1)} + for row in rows: for i in range(1, n_stages + 1): if row[f"stage{i}_timestamp"] is not None: users_in_stages[i].add(row["user_uuid"]) users_count = {i: len(users_in_stages[i]) for i in range(1, n_stages + 1)} - return users_count @@ -489,18 +483,18 @@ def get_issues(stages, rows, first_stage=None, last_stage=None, drop_only=False) last_stage = n_stages n_critical_issues = 0 - issues_dict = dict({"significant": [], - "insignificant": []}) + issues_dict = {"significant": [], + "insignificant": []} session_counts = count_sessions(rows, n_stages) drop = session_counts[first_stage] - session_counts[last_stage] - all_issues_with_context, n_issues_dict, affected_users_dict, affected_sessions, contexts = get_affected_users_for_all_issues( + all_issues, n_issues_dict, affected_users_dict, affected_sessions = get_affected_users_for_all_issues( rows, first_stage, last_stage) transitions, errors, all_errors, n_sess_affected = get_transitions_and_issues_of_each_type(rows, - all_issues_with_context, + all_issues, first_stage, last_stage) - # print("len(transitions) =", len(transitions)) + del rows if any(all_errors): total_drop_corr, conf, is_sign = pearson_corr(transitions, all_errors) @@ -513,33 +507,32 @@ def get_issues(stages, rows, first_stage=None, last_stage=None, drop_only=False) if drop_only: return total_drop_due_to_issues - for issue in all_issues_with_context: + for issue_id in all_issues: - if not any(errors[issue]): + if not any(errors[issue_id]): continue - r, confidence, is_sign = pearson_corr(transitions, errors[issue]) + r, confidence, is_sign = pearson_corr(transitions, errors[issue_id]) if r is not None and drop is not None and is_sign: - lost_conversions = int(r * affected_sessions[issue]) + lost_conversions = int(r * affected_sessions[issue_id]) else: lost_conversions = None if r is None: r = 0 - split = issue.split('__^__') issues_dict['significant' if is_sign else 'insignificant'].append({ - "type": split[0], - "title": helper.get_issue_title(split[0]), - "affected_sessions": affected_sessions[issue], - "unaffected_sessions": session_counts[1] - affected_sessions[issue], + "type": all_issues[issue_id]["issue_type"], + "title": helper.get_issue_title(all_issues[issue_id]["issue_type"]), + "affected_sessions": affected_sessions[issue_id], + "unaffected_sessions": session_counts[1] - affected_sessions[issue_id], "lost_conversions": lost_conversions, - "affected_users": affected_users_dict[issue], + "affected_users": affected_users_dict[issue_id], "conversion_impact": round(r * 100), - "context_string": contexts[issue]["context"], - "issue_id": contexts[issue]["id"] + "context_string": all_issues[issue_id]["context"], + "issue_id": issue_id }) if is_sign: - n_critical_issues += n_issues_dict[issue] + n_critical_issues += n_issues_dict[issue_id] return n_critical_issues, issues_dict, total_drop_due_to_issues