min_songs_hour = 10
output_listeningHistoryFolder = "../data/process/listeningHistory-hours/"
def doExperiment(lisHistFile, songsetFile):
df_lisHist = pd.read_csv(lisHistFile,delimiter=";", encoding="UTF-8") #"ISO-8859-1")
print(df_lisHist.head(1))
sys.exit()
contents_lh = glob(f"{listeningHistoryFolder}*.csv")
contents_lh.sort()
for listeningHistoryFile in contents_lh:
#df = pd.read_csv(listeningHistoryFile,delimiter=";", encoding="UTF-8")
#print(df.head())
df = mylib.loadData(listeningHistoryFile, min_msPlayed, delimiter="\t")
df.sort_values(by=["date","time","msPlayed"], inplace=True)
# Aggiungo una colonna con l'orario giusto
#df["MY_PLAYED_AT"] = [pd.Timestamp('2017-01-01T12') for i in range(0,df.shape[0])]
df["played_at"] = [" " for i in range(0,df.shape[0])]
df["datetime"] = [" " for i in range(0,df.shape[0])]
for index, row in df.iterrows():
# vado allo scadere del minuto e sottraggo i secondi
##df.at[index, "MY_PLAYED_AT"] = row["endTime"] + pd.Timedelta(seconds=(59-row["msPlayed"]//1000))
#df.at[index, "MY_PLAYED_AT"] = row["endTime"] + pd.Timedelta(seconds=59) - pd.Timedelta(milliseconds=row["msPlayed"])
played_at_timestamp = row["endTime"] + pd.Timedelta(seconds=59) - pd.Timedelta(milliseconds=row["msPlayed"])
# Problema: non stampa secondi e millisecondi se sono tutti 0
#timeiso = played_at_timestamp.isoformat()
#df.at[index, "played_at"] = timeiso[:-3] + "Z"
# In[] load 1M data
import mylib
[data_] = mylib.loadData('../1Mtrain')
train_data = data_[:900000]
valid_data = data_[900000:]
# In[] calculate with a lamda
from collections import defaultdict
from math import exp
import numpy
def getRui(data):
Rui = {};
for d in data:
Rui[(d['reviewerID'], d['itemID'])] = d['rating']
return Rui
def getIu_Ui(data):
Iu, Ui = {},{}
for d in data:
if d['reviewerID'] not in Iu:
Iu[d['reviewerID']]= []
if d['itemID'] not in Ui:
Ui[d['itemID']] = []
Iu[d['reviewerID']].append(d['itemID'])
Ui[d['itemID']].append(d['reviewerID'])
return [Iu, Ui]
Rui = getRui(train_data)
[Iu, Ui] = getIu_Ui(train_data)
def generateSongset(csv_file,
output_folder,
cluster_method="KM",
heuristic_method="LINEAR",
min_songs_hour=10,
min_ms_played=10000,
max_clusters=10,
num_tracks=100):
print(csv_file)
if ".csv" not in csv_file and ".tsv" not in csv_file:
print("ERROR: only tsv and csv input files are allowed. Skip file.")
cluster_method = cluster_method.upper()
heuristic_method = heuristic_method.upper()
df = mylib.loadData(csv_file, min_ms_played, delimiter="\t")
# 3.1
ntna_ntka = computeNTNA_NTKA(df)
for time_hour in range(0, 24):
if time_hour < 21:
continue
# 3.2 - FILTERING
df_h = songs_byHour(df, time_hour)
# Remove duplicate songs
df_h.drop_duplicates(subset="TrackID", keep="first", inplace=True)
# controllo su numero di canzoni nella fascia oraria
if df_h.shape[0] < min_songs_hour:
print(f"* hour {time_hour}: skip ({df_h.shape[0]} songs).")
continue
print(f"* hour {time_hour}: {df_h.shape[0]} songs.")
#3.3 - CLUSTERING
df_h_feat = df_h[[
"Acousticness", "Danceability", "Energy", "Instrumentalness",
"Key", "Liveness", "Loudeness", "Mode", "Speechiness", "Tempo",
"Time_signature", "Valence"
]]
if cluster_method == "KM":
best_clustering = best_k_means(
df_h_feat, max_clusters,
"exclude_K_less_4_songs") #"exclude_cluster_less_4_songs")
#print(best_clustering)
elif cluster_method == "FBF":
print(
f"ERROR in generateSongset(): {cluster_method} not yet implemented. Exit."
)
sys.exit()
else:
print(
f"ERROR in generateSongset(): cluster method {cluster_method} not defined. Exit."
)
sys.exit()
kLength = best_clustering["best-length"]
numReqs_perPoint = int(num_tracks / (4 * kLength)) + 1
feature_names = [
"Acousticness", "Danceability", "Energy", "Speechiness",
"Instrumentalness", "Liveness", "Valence", "Loudeness", "Tempo",
"Time_signature", "Key", "Mode"
]
########################
### LINEAR HEURISTIC ###
########################
if heuristic_method == "LINEAR":
linear_kMeans = linearHeuristic(df_h, best_clustering)
###########################
### RECOMMENDER SPOTIFY ###
###########################
results = list()
kIndex = 0
for df_group in linear_kMeans:
# FIRST SONG
print(
f"\t{time_hour}) CLUSTER KM #{kIndex}/{kLength} - SONG #0/4"
)
firstPoint = df_group.iloc[0]
trackId = firstPoint["TrackID"]
# get features
centroidFeatures_list = firstPoint["kCentroid"]
features = dict()
for index in range(0, len(centroidFeatures_list)):
features[
feature_names[index]] = centroidFeatures_list[index]
tracks = recommenderGetSongs(trackId,
features,
numReqs_perPoint,
results,
retryLimit=2,
sleepTime=recommender_sleepTime)
results.extend(tracks)
#print(len(results))
#res = recommenderGetSongs("7CDaY0pk8qGFoahgxVVbaX", numReqs_perPoint, list(), retryLimit=2, sleepTime=recommender_sleepTime)
# OTHER THREE SONGS
for i in range(1, 4):
print(
f"\t{time_hour}) CLUSTER KM LINEAR #{kIndex}/{kLength} - SONG #{i}/4"
)
point = df_group.iloc[i]
trackId = point["TrackID"]
tracks = recommenderGetSongs(
trackId,
point,
numReqs_perPoint,
results,
retryLimit=2,
sleepTime=recommender_sleepTime)
results.extend(tracks)
#print(len(results))
kIndex += 1
########################
### SPHERE HEURISTIC ###
########################
elif heuristic_method == "SPHERE":
sphere_kMeans = sphereHeuristic(df_h, best_clustering)
###########################
### RECOMMENDER SPOTIFY ###
###########################
results = list()
for item in sphere_kMeans:
print(
f"\t{time_hour}) {kLength} CLUSTER KM SPHERE - SONG #{item['index']}/{len(sphere_kMeans)}"
)
#{"index": minDistSongIndex, "randomPoint": currRandomPoint, "minDistSong": minDistSong, "minDist": minDist}
randomPoint = item["randomPoint"]
features = dict()
for index in range(0, len(randomPoint)):
features[feature_names[index]] = randomPoint[index]
#for index in range(0,len(centroidFeatures_list)):
# features[feature_names[index]] = centroidFeatures_list[index]
clusterMinDistSong = item["minDistSong"]
trackId = clusterMinDistSong["TrackID"]
tracks = recommenderGetSongs(trackId,
features,
numReqs_perPoint,
results,
retryLimit=2,
sleepTime=recommender_sleepTime)
results.extend(tracks)
else:
print(
f"ERROR in generateSongset(): heuristic {cluster_method} not defined. Exit."
)
sys.exit()
#print(results)
output_file_start = ntpath.basename(csv_file).replace(".csv",
"").replace(
".tsv", "")
saveSongset(results, output_folder, output_file_start, time_hour,
kLength, cluster_method, heuristic_method)
pairs_Rating.append(UIO)
rating_test_predict = []
for pr in pairs_Rating:
if pr[0] in dirty_u.keys():
rating_test_predict.append(dirty_u[pr[0]])
elif pr[1] in dirty_i.keys():
rating_test_predict.append(dirty_i[pr[1]])
else:
rating_test_predict.append(alpha+beta_u[pr[0]]+beta_i[pr[1]])
rating_test_result = [[pr[0]+'-'+pr[1], str(ptr)] for pr,ptr in zip(pairs_Rating,rating_test_predict)]
saveCSV('rating_test_result.csv',rating_test_result)
# In[]================================================================================
# In[] main
[data_] = mylib.loadData('./assignment1/data/1Mtrain')
train_data = data_[:900000]
valid_data = data_[900000:]
del data_
dirty_limits = [10,11,12,13,14,15,16,17,18,19,20]
sds = [0.3,0.5,0.7,0.9,1.0,2.0]
dirty_bound = [[2.6,4.4],[2.8,4.2],[3,4],[3.5,3.5]]
MSEs = [[[0 for dbi in range(len(dirty_bound))] for si in range(len(sds))] for dli in range(len(dirty_limits))]
thetas = [[[0 for dbi in range(len(dirty_bound))] for si in range(len(sds))] for dli in range(len(dirty_limits))]
for dli in range(len(dirty_limits)):
dl = dirty_limits[dli]
for si in range(len(sds)):
s = sds[si]
for dbi in range(len(dirty_bound)):
db = dirty_bound[dbi]
"""
Created on Thu Nov 12 22:56:19 2015
@author: ssc317
"""
import mylib
import csv
filename = '1M_train_rating'
pairs_Rating = []
f = open('./pairs_Rating.txt')
for line in f:
if line.startswith('userID'):
pass
elif line.startswith('U'):
UIO = line.split('-')
pairs_Rating.append(UIO)
[alpha, beta_u, beta_i] = mylib.loadData(filename)
rating_test_predict = [alpha+beta_u[pr[0]]+beta_i[pr[1]] for pr in pairs_Rating]
rating_test_result = [[pr[0]+'-'+pr[1].rstrip(), str(ptr)] for pr,ptr in zip(pairs_Rating,rating_test_predict)]
def saveCSV(filename, data):
f = open(filename, 'wb')
writer = csv.writer(f)
writer.writerow(['userID-itemID', 'prediction'])
for d in data:
writer.writerow(d)
f.close()
saveCSV('rating_test_result.csv',rating_test_result)
