def count_comments_likes(months=12):
'''
Returns:
* total number of comments
* total number of likes
* number of photos which had 0 comments
* number of photos which had 0 likes
* number of photos which had > 0 comments
* number of photos which had > 0 likes
'''
img_usr = __data.load_image_data(months)[["user_id", "image_id"]]
comments_likes = __img_f.final_like_and_comments()
com_l1 = comments_likes.comments > 0
com_l2 = comments_likes.comments == 0
like_l1 = comments_likes.likes > 0
like_l2 = comments_likes.likes == 0
comments_likes["num_images_with_comments"] = __np.where(com_l1, 1, 0)
comments_likes["num_images_no_comments"] = __np.where(com_l2, 1, 0)
comments_likes["num_images_with_likes"] = __np.where(like_l1, 1, 0)
comments_likes["num_images_no_likes"] = __np.where(like_l2, 1, 0)
out = img_usr.merge(comments_likes, on="image_id", how="left").drop("image_id",axis=1)\
.groupby("user_id").sum().reset_index()
out = out.rename(columns={
"likes": "total_number_of_likes",
"comments": "total_number_of_comments"
})
return out
def ratio_of_topics(confidence=80, subset=True, months=12):
'''
NOTE: This is an expensive operation.
Returns a matrix of the percentages that shows you what percentage that object appears in the users photos.
Arg:
confidence: confidence of topic being in image
subset: returns subset of topics for the user that we selected in advance as indicators of lifestyle
and which were not sparse.
'''
image_data = __data.load_image_data(months)
#__pd.read_pickle("../../data/Visual_well_being/image_data.pickle")
user_img = image_data[["image_id", "user_id"]].drop_duplicates()
ob = __img_f.binary_object_matrix(confidence)
photo_counts = instagram_account_stats()[["user_id", "user_posted_photos"]]
counts_per_user = ob.merge(user_img,how="right",on="image_id")\
.groupby("user_id").sum()\
.merge(photo_counts,on="user_id",how="inner")
f = counts_per_user.set_index("user_id")
x = (f.sum() > 30)
df = f.iloc[:, list(x.values)]
cols = []
for i in df.columns:
if i != "user_posted_photos":
cols.append("ratio_" + i)
else:
cols.append(i)
df.columns = cols
df = df.reset_index()
out = df.iloc[:,1:-1]\
.divide(
df.user_posted_photos,
axis=0
)
out["user_id"] = df.user_id
#if subset == True:
# topics_considered = ["user_id","Person","Plant","Food","Collage","Animal","Outdoors","Pet","Book","Dog","Canine","Sky","Alcohol","Crowd","Toy","Cat","Coast","Tree","Beach","Sport","Teddy Bear","Sunlight","Light","Drawing","Sea Life","TV","Dusk","Bikini","Sunrise","Sunset","Swimwear","Selfie","Beard","Woman","Cocktail","Pool","Performer","Coffee Cup","Tattoo","Downtown","Musical Instrument","Festival","City","Laptop","Pizza","Cloud","Beer Bottle","Money","Club","Airplane","Sketch","Sandwich","Cafeteria","Breakfast","Child"]
# df = df[topics_considered]
return out
def avg_posts_per_day(months=12):
'''
Returns the average number of posts per day that the person posted. Returns 7 averages:
`early day`: 8:00-12:00
`late_day`: 12:00-20:00
`early_night`: 20:00-00:00
`late_night`: 00:00-8:00
`day`: 8:00-20:00
`night`: 20:00-8:00
`whole_day`: the average for the whole date
'''
image_data = __data.load_image_data(months)
#__pd.read_pickle(__data_dir + "image_data.pickle")
#image_date.image_posted_time = __pd.to_datetime(image_date.image_posted_time)
x = image_data[["image_posted_time", "image_id",
"user_id"]].drop_duplicates()
early_day_b = x.image_posted_time.isin(
x.set_index("image_posted_time").between_time('8:00', '12:00').index)
late_day_b = x.image_posted_time.isin(
x.set_index("image_posted_time").between_time('12:00', '20:00').index)
early_night_b = x.image_posted_time.isin(
x.set_index("image_posted_time").between_time('20:00', '00:00').index)
late_night_b = x.image_posted_time.isin(
x.set_index("image_posted_time").between_time('00:00', '8:00').index)
day_b = x.image_posted_time.isin(
x.set_index("image_posted_time").between_time('8:00', '20:00').index)
night_b = x.image_posted_time.isin(
x.set_index("image_posted_time").between_time('20:00', '8:00').index)
x["avg_posts_early_day"] = __np.where(early_day_b, 1, 0)
x["avg_posts_late_day"] = __np.where(late_day_b, 1, 0)
x["avg_posts_early_night"] = __np.where(early_night_b, 1, 0)
x["avg_posts_late_night"] = __np.where(late_night_b, 1, 0)
x["avg_posts_day"] = __np.where(day_b, 1, 0)
x["avg_posts_night"] = __np.where(night_b, 1, 0)
x["avg_posts_whole_date"] = 1
xx = x.drop(columns="image_id",axis=0)\
.groupby([x.user_id,x.image_posted_time.dt.date])\
.sum().drop("user_id",axis=1).reset_index()
out = xx.groupby("user_id").mean().reset_index()
return out
def avg_ratio_gender(confidence=90, months=12):
'''
Returns the average of the gender ratio that the user has per image
'''
image_data = __data.load_image_data(months)
#__pd.read_pickle(__data_dir + "image_data.pickle")
ratio_gender = __img_f.ratio_gender(confidence)
avg_ratio_gender = __pd.merge(ratio_gender,
image_data[['user_id', 'image_id']],
on='image_id',
how='right')
avg_ratio_gender.fillna(0, inplace=True)
return avg_ratio_gender.groupby('user_id').mean().reset_index()
def average_num_faces_per_image_and_emotion(months=12):
'''
Returns the average of faces per emotion that the user has per image
'''
image_data = __data.load_image_data(months)
#__pd.read_pickle(__data_dir + "image_data.pickle")
num_faces_df = __img_f.number_of_faces_per_emotion()
num_faces_df = __pd.merge(num_faces_df,
image_data[['user_id', 'image_id']],
on='image_id',
how='right')
num_faces_df.fillna(0, inplace=True)
return num_faces_df.groupby('user_id').mean().reset_index()
def instagram_account_stats():
'''
Returns statistics on the instagram user.
* Number of followers
* Number of people following
* Number of photos
'''
image_data = __data.load_image_data()
instagram_account_info = image_data[["user_id","user_followed_by","user_follows","user_posted_photos"]]\
.drop_duplicates()
return instagram_account_info
def avg_number_of_faces_from_photos_with_faces(months=12):
'''Returns the average number of faces in photos with faces'''
image_data = __data.load_image_data(months)
#__pd.read_pickle("../../data/Visual_well_being/image_data.pickle")
num_faces = __img_f.number_of_faces()
out = image_data[["image_id","user_id"]]\
.merge(num_faces,on="image_id",how="left")\
.fillna(0)\
.groupby("user_id")\
.mean()\
.rename(columns = {'number_of_face': 'avg_number_of_faces_over_images_with_faces'})\
.reset_index()
return out
def count_face_emotions(months=12):
emotion = __img_f.number_of_faces_per_emotion()
img_usr = __data.load_image_data(months=12)[["user_id", "image_id"]]
out = img_usr.merge(emotion,on="image_id",how="inner").drop(["image_id"],axis=1)\
.groupby("user_id").sum().reset_index()
out = out.rename(
columns={
"ANGRY": "count_ANGRY",
"CALM": "count_CALM",
"CONFUSED": "count_CONFUSED",
"DISGUSTED": "count_DISGUSTED",
"HAPPY": "count_HAPPY",
"SAD": "count_SAD",
"SUPRISED": "count_SUPRISED"
})
return out
def avg_engagement(months=12):
'''
Returns the average number of likes and comments per user.
'''
image_data = __data.load_image_data(months)
#__pd.read_pickle('../../data/Visual_well_being/image_data.pickle')
updated_metrics = __img_f.final_like_and_comments(months)
image_data = image_data.merge(updated_metrics, how='left', on='image_id')
avg_engagement = image_data[['user_id', 'likes', 'comments'
]].groupby('user_id').mean().reset_index()
avg_engagement = avg_engagement.rename(columns={
'likes': 'avg_likes',
'comments': 'avg_comments'
})
return avg_engagement
def counts_objects(months=12, confidence=80):
o = __img_f.binary_object_matrix(confidence)
image_data = __data.load_image_data(months)
user_img = image_data[["image_id", "user_id"]].drop_duplicates()
out = o.merge(user_img,how="right",on="image_id")\
.groupby("user_id").sum()\
.fillna(0)
x = (out.sum() > 30)
out = out.iloc[:, list(x.values)]
cols = []
for i in out.columns:
cols.append("count_" + i)
out.columns = cols
out = out.reset_index()
return out
def proportion_image_cluster(months=12):
anp_cg = __img_f.anp_cluster_groups()
u = instagram_account_stats()[["user_id", "user_posted_photos"]]
image_data = __data.load_image_data(months)
image_user = image_data[["image_id", "user_id"]]
user_clusters = image_user.merge(anp_cg, on="image_id",
how="left").fillna(0).drop("image_id",
axis=1)
user_clusters = user_clusters.groupby("user_id").sum()\
.reset_index()\
.merge(u,on="user_id",how="inner")\
.reset_index()
cluster_proportions = __pd.concat([user_clusters.user_id,
user_clusters.iloc[:,2:-1]\
.divide(user_clusters.user_posted_photos,
axis=0)
],axis=1)
return cluster_proportions
def filter_features(months=12):
'''
Returns percentage of happy/depressed filters, ratio of happy over depressed filters
'''
image_data = __data.load_image_data(months)
#__pd.read_pickle('../../data/Visual_well_being/image_data.pickle')
# Keep only filters, remove Unknown and Normal entries
filter_data = image_data[~image_data.image_filter.
isin(['Normal', 'Unknown'])][[
'image_id', 'image_filter'
]]
# Load the filter categories
filter_categories = __pd.read_csv(
'../../data/Visual_well_being/filter_categories.csv', sep=';')
filter_categories = filter_categories.rename(
columns={'class': 'happiness_class'})
# Remove images whose filter is not associated with a category in filter_categories (only 23 of them, no big deal)
filter_data = filter_data[filter_data.image_filter.isin(
filter_categories['filter'])]
# Add filter category information to the dataFrame
filter_data = filter_data.merge(filter_categories,
how='left',
left_on='image_filter',
right_on='filter').drop('filter', axis=1)
# Create Dummies that will help to summarize happy filters and depressed filters later on.
filter_dummies = __pd.get_dummies(
filter_data['happiness_class']).rename(columns={
0: 'depressed_filter',
1: 'happy_filter'
})
filter_data['happy_filter'] = filter_dummies['happy_filter']
filter_data['depressed_filter'] = filter_dummies['depressed_filter']
filter_data = filter_data.drop(['image_filter', 'happiness_class'], axis=1)
# Merge with original image data
image_data = image_data.merge(filter_data, 'left', 'image_id')
# Create Filter features dataframe
filter_features = image_data[[
'user_id', 'happy_filter', 'depressed_filter'
]].groupby('user_id').sum().reset_index()
filter_features['total_photos'] = image_data[[
'user_id', 'user_posted_photos'
]].groupby('user_id').max().reset_index()['user_posted_photos']
# Build the features
filter_features['happy_flt_pct'] = filter_features[
'happy_filter'] / filter_features['total_photos']
filter_features['depressed_flt_pct'] = filter_features[
'depressed_filter'] / filter_features['total_photos']
filter_features['happy_to_depressed_flt_ratio'] = filter_features[
'happy_filter'] / filter_features['depressed_filter']
filter_features = filter_features.replace([np.inf, -np.inf], np.nan)
# Drop not neeed columns
filter_features = filter_features.drop(
['happy_filter', 'depressed_filter', 'total_photos'], axis=1)
return filter_features
