def calcItemInfo(item, user_ratings, users_info, alpha):
'''Qi, Lii(i) for adjusted cosine'''
Q = 0
L = 0
n = 0 # item popularity
for rating in user_ratings:
user = rating[0]
ri = rating[1]
ti = rating[2]
t, avg, e = users_info[user]
Q += (f(t, ti, alpha) * (ri - avg))**2
L += f(t, ti, alpha)**2 * (ri - avg) * e
n += 1
return item, L, n, Q
def updateItemInfo(iid, prevL, prevn, prevQ, user_t,
prev_ratings, fresh_ratings, rubar, user_e, alpha):
'''Calculate the incrementaly update value of Qi and Lii(i)'''
"""calculate Q, L"""
dL = 0
dQ = 0
dn = 0
for (user, item) in fresh_ratings:
if item == iid: # u\epsilon\DeltaU_i^t
rating, timestep = fresh_ratings[(user, item)]
ru = rubar[user]
e = user_e[user]
t = user_t[user]
dL += (rating - ru) * e
dQ += (rating - ru)**2
dn += 1
if (user, item) in prev_ratings: # u\epsilon\U_i^{t-1}
dL += 2 * f(t, timestep, alpha) * (rating - ru) * e
L = dL + math.exp(-2*alpha) * prevL
Q = dQ + math.exp(-2*alpha) * (prevQ - 2 * prevL)
n = dn + prevn
if Q < 0:
Q = prevQ
L = prevL
return iid, L, n, Q
def compute_Sapprox(csv_file):
starting_state, data = preprocessing_data(csv_file)
#starting_state = np.array(pd.read_csv('startingstate.csv').iloc[:,1])
N = len(starting_state)
Sapprox = simulated_annealing(starting_state,
betas,
n_iter,
lambda_,
data,
verbose=True,
plot=True)
#Computing score of this run with the Sapprox found and save it in a file called Aquarium_submission_SCORE with SCORE being the score of the run.
score = f(np.array(Sapprox), lambda_, data)
submission(csv_file, Sapprox,
"Aquarium_submission_N_{}_score_{}.csv".format(N, score))
def updateItemPairInfo(i, j, prevLi, prevLj, prevP, fresh_users, user_t,
prev_ratings, fresh_ratings, rubar, user_e, alpha):
'''Calculate the incrementaly update value of Pij, Lij(i), Lij(j)'''
"""calculate Pij, Lij(i), Lij(j)"""
dP = 0
dLi = 0
dLj = 0
for user in fresh_users:
ru = rubar[user]
e = user_e[user]
t = user_t[user]
# u\epsilon\DeltaU_i^t \cap u\epsilon\U_j^{t-1}
if (user, i) in fresh_ratings and (user, j) in prev_ratings:
isRated = True
ri, ti = fresh_ratings[(user, i)]
rj, tj = prev_ratings[(user, j)]
dP += (ri - ru) * f(t, tj, alpha) * (rj - ru)
dLi += f(t, tj, alpha) * (ri - ru) * e
dLj += f(t, tj, alpha) * (rj - ru) * e
# u\epsilon\DeltaU_j^t \cap u\epsilon\U_i^{t-1}
if (user, j) in fresh_ratings and (user, i) in prev_ratings:
isRated = True
rj, tj = fresh_ratings[(user, j)]
ri, ti = prev_ratings[(user, i)]
dP += f(t, ti, alpha) * (ri - ru) * (rj - ru)
dLi += f(t, ti, alpha) * (ri - ru) * e
dLj += f(t, ti, alpha) * (rj - ru) * e
# u\epsilon\DeltaU_i^t \cap u\epsilon\DeltaU_j^t
if (user, i) in fresh_ratings and (user, j) in fresh_ratings:
isRated = True
ri, ti = fresh_ratings[(user, i)]
rj, tj = fresh_ratings[(user, j)]
dP += (ri - ru) * (rj - ru)
dLi += (ri - ru) * e
dLj += (rj - ru) * e
Li = dLi + math.exp(-2*alpha) * prevLi
Lj = dLj + math.exp(-2*alpha) * prevLj
P = dP + math.exp(-2*alpha) * (prevP - prevLi - prevLj)
return i, j, Li, Lj, P
def calcItemPairInfo(itemi, itemj, rating_pairs, user_info, alpha):
'''For each i-j item pair, return Pij, Lij(i), Lij(j)'''
Pij = 0
Lij_i = 0
Lij_j = 0
for (user, i, j) in rating_pairs:
t, avg, e = user_info[user]
ri = np.float(i[0])
ti = i[1]
rj = np.float(j[0])
tj = j[1]
Pij += f(t, ti, alpha) * (ri - avg) * f(t, tj, alpha) * (rj - avg)
Lij_i += f(t, ti, alpha) * f(t, tj, alpha) * (ri - avg) * e
Lij_j += f(t, ti, alpha) * f(t, tj, alpha) * (rj - avg) * e
return itemi, itemj, Lij_i, Lij_j, Pij