# Time-series impressions (DOW, HOD, etc) (0 = Sunday... 6 = Saturday)
gmt_offset = -4

# Create proper datetime column, apply local GMT offset
tweet_df['ts'] = pd.to_datetime(tweet_df['time'])
tweet_df['ts'] = tweet_df.ts + pd.to_timedelta(gmt_offset, unit='h')

# Add hour of day and day of week columns
tweet_df['hod'] = [t.hour for t in tweet_df.ts]
tweet_df['dow'] = [t.dayofweek for t in tweet_df.ts]

hod_dict = {}
hod_count = {}
dow_dict = {}
dow_count = {}
weekday_dict = {0: 'Mon', 1: 'Tue', 2: 'Wed', 3: 'Thu', 4: 'Fri', 5: 'Sat', 6: 'Sun'}

# Process tweets, collect stats
for i in tweet_df.index:
    hod = tweet_df.ix[i]['hod']
    dow = tweet_df.ix[i]['dow']
    imp = tweet_df.ix[i]['impressions']

    if hod in hod_dict:
        hod_dict[hod] += int(imp)
        hod_count[hod] += 1
    else:
        hod_dict[hod] = int(imp)
        hod_count[hod] = 1

    if dow in dow_dict:
        dow_dict[dow] += int(imp)
        dow_count[dow] += 1
    else:
        dow_dict[dow] = int(imp)
        dow_count[dow] = 1

print 'Average impressions per tweet by hour tweeted:'
print '----------------------------------------------'
for hod in hod_dict:
    print hod, '-', hod+1, ':', hod_dict[hod]/hod_count[hod], '=>', hod_count[hod], 'tweets'

print '\nAverage impressions per tweet by day of week tweeted:'
print '-----------------------------------------------------'
for dow in dow_dict:
    print weekday_dict[dow], ':', dow_dict[dow]/dow_count[dow], '=>', dow_count[dow], ' tweets'