def get_userdf_parts(user_df):
"""
Parse user profile part of json
user_df (dict):
Raw dictionary from json that contains user information
Returns
user_df_kept (dict)
Subset of user_df with keys of interest
Parses user file... this section contains a lot of person information which I purposely try to avoid
"""
keep_keys = [
'active_time', 'age_filter_max', 'age_filter_min', 'bio', 'birth_date',
'connection_count', 'create_date', 'education', 'gender',
'gender_filter', 'jobs', 'schools'
]
for key in keep_keys:
if key not in user_df:
raise IndexError(key + " not in input keys: " +
str(user_df.keys()))
user_df_kept = {k: v for k, v in user_df.items() if k in keep_keys}
user_df_kept['birth_date'] = user_df_kept['birth_date'][:10]
user_df_kept = utils.check_dict_types(user_df_kept)
return (user_df_kept)
def gather_usage_stats(usage_df):
usage_sums = usage_df.sum()
derived_metrics = {}
# Number of swipes
derived_metrics["total_swipes"] = usage_sums['swipes_likes'] + usage_sums[
'swipes_passes']
# Likes to passes (for every 1 pass there are x likes )
derived_metrics["like_to_pass_ratio"] = usage_sums[
'swipes_likes'] / usage_sums['swipes_passes']
# Number of swipes per app open
derived_metrics["swipes/app_open"] = (
usage_sums['swipes_likes'] +
usage_sums['swipes_passes']) / usage_sums['app_opens']
# Avg Messages recieved per match
derived_metrics["n_avg_msg_rec_per_match"] = usage_sums[
'messages_received'] / usage_sums['matches']
# Avg Messages sent per match
derived_metrics["n_avg_msg_sent_per_match"] = usage_sums[
'messages_sent'] / usage_sums['matches']
# Calendar days on tinder
days_on_tinder = pd.to_datetime(usage_df.iloc[-1].name) - pd.to_datetime(
usage_df.iloc[0].name)
days_on_tinder = days_on_tinder.days
# Days when you opened the app
active_days_on_tinder = len(usage_df.index)
# Avg Swipes per day
derived_metrics["swipes_per_tot_cal_day"] = derived_metrics[
"total_swipes"] / days_on_tinder
derived_metrics["swipes_per_act_day"] = derived_metrics[
"total_swipes"] / active_days_on_tinder
derived_metrics = {k: np.round(v, 2) for k, v in derived_metrics.items()}
derived_metrics = utils.check_dict_types(derived_metrics)
return (derived_metrics)
def get_message_metrics(message_df):
"""
Get metrics from message dataframe
message_df (pandas DataFrame):
Dataframe with info about messages sent, time, content of message, and whom it was sent
Returns
metrics_to_save (dict):
dictionary of name of metric and value
"""
# Input Validation
expected_columns = ['sent_date', 'n_words_in_msg', 'message']
for col in expected_columns:
if col not in message_df.columns:
raise IndexError(col + " not in dataframe columns " +
message_df.columns)
expected_index_keys = ['match_id', 'msg_number']
for col in expected_index_keys:
if col not in message_df.index.names:
raise IndexError(col + " not in dataframe columns " +
message_df.index.names)
# Set up -- needed for multi index slicing
idx = pd.IndexSlice
# Metrics to save
metrics_to_save = {}
metrics_to_save["Date of First Message Sent"] = message_df[
'sent_date'].min().strftime('%b %d %Y')
metrics_to_save["Date of Last Message Sent"] = message_df['sent_date'].max(
).strftime('%b %d %Y')
metrics_to_save["Number of Matches"] = message_df.index.get_level_values(
'match_id').nunique()
metrics_to_save["Number Matches with no Messages"] = message_df.loc[
idx[:, -1], ].shape[0]
metrics_to_save[
"Most Number of Messages Sent to a Match"] = message_df.index.get_level_values(
'msg_number').max()
metrics_to_save["Average Number of Words per Message"] = message_df[
'n_words_in_msg'].mean().round(2)
metrics_to_save["Median Number of Words per Message"] = np.round(
message_df['n_words_in_msg'].quantile(0.50), 2)
# metrics_to_save["Average Messages Per Match"] = message_df.index.get_level_values('msg_number').mean().round(3)
# Time calculations
time_diff_days = (message_df['sent_date'].max() -
message_df['sent_date'].min()).days
years = int(time_diff_days / 365)
months = int((time_diff_days % 365) / 30)
days = (time_diff_days % 365) % 30
metrics_to_save["Total Time on Tinder from First Message to Last"] = "{} years {} months {} days" \
.format(years, months, days)
# First message analysis
first_msg = message_df.loc[idx[:, 0], "message"]
first_msg = pd.DataFrame(first_msg)
# Clean and generate additional flags
first_msg['message'] = first_msg['message'].str.rstrip(' ')
first_msg['message'] = first_msg['message'].str.rstrip(' ')
same_first_message = first_msg['message'].value_counts().sort_values(
ascending=False)
first_msg['hey_hi_flag'] = first_msg['message'].str.count(
'((H|h)ey|(H|h)i)')
first_msg['How \'sit going_flag'] = first_msg['message'].str.contains(
'it going')
# Save some metrics
metrics_to_save["Hey or Hi in First Message"] = first_msg[
'hey_hi_flag'].sum()
metrics_to_save["(How's ) it going in First Message"] = first_msg[
'How \'sit going_flag'].sum()
metrics_to_save["Most Common First Message"] = same_first_message.index[0]
metrics_to_save[
"Number of Times Most Common First Message Used"] = same_first_message[
0]
metrics_to_save[
"Second Most Common First Message"] = same_first_message.index[1]
metrics_to_save[
"Number of Times Second Most Common First Message Used"] = same_first_message[
1]
metrics_to_save[
"Third Most Common First Message"] = same_first_message.index[2]
metrics_to_save[
"Number of Times Third Most Common First Message Used"] = same_first_message[
2]
metrics_to_save = utils.check_dict_types(metrics_to_save)
return (metrics_to_save)