alu/CSE-158-Assignment-1
Archived
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

use BPR for predict play

Browse Source
This commit is contained in:
ltcptgeneral 2023-11-04 01:10:01 -07:00
parent c760ea5e14
commit aa423db398
3 changed files with 3971 additions and 3580 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

196
assignment1.ipynb
View File

@ -9,19 +9,20 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 27, "execution_count": 38,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
"# load data into dataset array\n", "# load data into dataset array\n",
"import gzip\n", "import gzip\n",
"from collections import defaultdict\n", "from collections import defaultdict\n",
"import numpy as np" "import numpy as np\n",
"import tensorflow as tf"
] ]
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 28, "execution_count": 39,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@ -44,7 +45,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 29, "execution_count": 40,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@ -59,15 +60,14 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 30, "execution_count": 41,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
"# Get negative labels in vaidation\n", "# Get negative labels in vaidation\n",
"import random\n", "import random\n",
"\n", "\n",
"def get_balanced_validation (dataset, valid):\n", "def get_balanced_set(dataset, s):\n",
"\n",
" all_games = set()\n", " all_games = set()\n",
" user_played = defaultdict(set)\n", " user_played = defaultdict(set)\n",
"\n", "\n",
@ -75,14 +75,15 @@
" all_games.add(review[\"gameID\"])\n", " all_games.add(review[\"gameID\"])\n",
" user_played[review[\"userID\"]].add(review[\"gameID\"])\n", " user_played[review[\"userID\"]].add(review[\"gameID\"])\n",
"\n", "\n",
" negative_valid = []\n", " negative = []\n",
"\n", "\n",
" for user,game,review in valid:\n", " for user,game,review in s:\n",
" not_played = all_games - user_played[user]\n", " not_played = all_games - user_played[user]\n",
" new_game = random.choice(tuple(not_played))\n", " new_game = random.choice(tuple(not_played))\n",
" negative_valid.append((user, new_game, {\"played\": 0}))\n", " negative.append((user, new_game, {\"played\": 0}))\n",
"\n", "\n",
" return valid + negative_valid" " return s + negative\n",
" "
] ]
}, },
{ {
@ -94,7 +95,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 31, "execution_count": 42,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@ -113,6 +114,48 @@
" predictions.close()" " predictions.close()"
] ]
}, },
{
"cell_type": "code",
"execution_count": 43,
"metadata": {},
"outputs": [],
"source": [
"class BPRbatch(tf.keras.Model):\n",
" def __init__(self, K, lamb, itemIDs, userIDs):\n",
" super(BPRbatch, self).__init__()\n",
" # Initialize variables\n",
" self.betaI = tf.Variable(tf.random.normal([len(itemIDs)],stddev=0.001))\n",
" self.gammaU = tf.Variable(tf.random.normal([len(userIDs),K],stddev=0.001))\n",
" self.gammaI = tf.Variable(tf.random.normal([len(itemIDs),K],stddev=0.001))\n",
" # Regularization coefficient\n",
" self.lamb = lamb\n",
"\n",
" # Prediction for a single instance\n",
" def predict(self, u, i):\n",
" p = self.betaI[i] + tf.tensordot(self.gammaU[u], self.gammaI[i], 1)\n",
" return p\n",
"\n",
" # Regularizer\n",
" def reg(self):\n",
" return self.lamb * (tf.nn.l2_loss(self.betaI) +\\\n",
" tf.nn.l2_loss(self.gammaU) +\\\n",
" tf.nn.l2_loss(self.gammaI))\n",
" \n",
" def score(self, sampleU, sampleI):\n",
" u = tf.convert_to_tensor(sampleU, dtype=tf.int32)\n",
" i = tf.convert_to_tensor(sampleI, dtype=tf.int32)\n",
" beta_i = tf.nn.embedding_lookup(self.betaI, i)\n",
" gamma_u = tf.nn.embedding_lookup(self.gammaU, u)\n",
" gamma_i = tf.nn.embedding_lookup(self.gammaI, i)\n",
" x_ui = beta_i + tf.reduce_sum(tf.multiply(gamma_u, gamma_i), 1)\n",
" return x_ui\n",
"\n",
" def call(self, sampleU, sampleI, sampleJ):\n",
" x_ui = self.score(sampleU, sampleI)\n",
" x_uj = self.score(sampleU, sampleJ)\n",
" return -tf.reduce_mean(tf.math.log(tf.math.sigmoid(x_ui - x_uj)))"
]
},
{ {
"cell_type": "markdown", "cell_type": "markdown",
"metadata": {}, "metadata": {},
@ -122,7 +165,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 32, "execution_count": 44,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@ -131,10 +174,59 @@
" def __init__(self):\n", " def __init__(self):\n",
" pass\n", " pass\n",
"\n", "\n",
" def fit(self, data, threshold=0.6): # data is an array of (user, game, review) tuples\n", " def fit(self, data, threshold=0.6, K=5, iters=100): # data is an array of (user, game, review) tuples\n",
" self.topGames = self.getTopGames(threshold)\n", " self.topGames = self.getTopGames(threshold)\n",
"\n", "\n",
" def predict(self, user, game):\n", " self.userIDs = {}\n",
" self.itemIDs = {}\n",
" interactions = []\n",
"\n",
" for u,i,r in data:\n",
" if not u in self.userIDs: self.userIDs[u] = len(self.userIDs)\n",
" if not i in self.itemIDs: self.itemIDs[i] = len(self.itemIDs)\n",
" interactions.append((u,i,r[\"played\"]))\n",
" \n",
" items = list(self.itemIDs.keys())\n",
" \n",
" itemsPerUser = defaultdict(list)\n",
" usersPerItem = defaultdict(list)\n",
" for u,i,r in interactions:\n",
" itemsPerUser[u].append(i)\n",
" usersPerItem[i].append(u)\n",
"\n",
" def trainingStepBPR(model, interactions):\n",
" Nsamples = 50000\n",
" with tf.GradientTape() as tape:\n",
" sampleU, sampleI, sampleJ = [], [], []\n",
" for _ in range(Nsamples):\n",
" u,i,_ = random.choice(interactions) # positive sample\n",
" j = random.choice(items) # negative sample\n",
" while j in itemsPerUser[u]:\n",
" j = random.choice(items)\n",
" sampleU.append(self.userIDs[u])\n",
" sampleI.append(self.itemIDs[i])\n",
" sampleJ.append(self.itemIDs[j])\n",
"\n",
" loss = model(sampleU,sampleI,sampleJ)\n",
" loss += model.reg()\n",
" gradients = tape.gradient(loss, model.trainable_variables)\n",
" optimizer.apply_gradients((grad, var) for\n",
" (grad, var) in zip(gradients, model.trainable_variables)\n",
" if grad is not None)\n",
" return loss.numpy()\n",
" \n",
" optimizer = tf.keras.optimizers.Adam(0.1)\n",
" self.modelBPR = BPRbatch(K, 0.00001, self.itemIDs, self.userIDs)\n",
"\n",
" for i in range(iters):\n",
" obj = trainingStepBPR(self.modelBPR, interactions)\n",
" if (i % 10 == 9): print(\"iteration \" + str(i+1) + \", objective = \" + str(obj))\n",
" \n",
" def predict(self, user, game, threshold=0.5):\n",
" if user in self.userIDs and game in self.itemIDs:\n",
" pred = self.modelBPR.predict(self.userIDs[user], self.itemIDs[game]).numpy()\n",
" return int(pred > threshold)\n",
" else:\n",
" return int(game in self.topGames)\n", " return int(game in self.topGames)\n",
"\n", "\n",
" def getTopGames (self, threshold):\n", " def getTopGames (self, threshold):\n",
@ -155,37 +247,50 @@
" count += ic\n", " count += ic\n",
" return1.add(i)\n", " return1.add(i)\n",
" if count > totalPlayed * threshold: break\n", " if count > totalPlayed * threshold: break\n",
" return return1" " return return1\n"
] ]
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 33, "execution_count": 45,
"metadata": {},
"outputs": [],
"source": [
"model = PlayPredictor()\n",
"model.fit(train, threshold=0.6)"
]
},
{
"cell_type": "code",
"execution_count": 34,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{ {
"name": "stdout", "name": "stdout",
"output_type": "stream", "output_type": "stream",
"text": [ "text": [
"PlayPredictor accuracy: 0.6997699769976997\n" "iteration 10, objective = 0.51180786\n",
"iteration 20, objective = 0.48082852\n",
"iteration 30, objective = 0.47100148\n",
"iteration 40, objective = 0.45862892\n",
"iteration 50, objective = 0.45290428\n",
"iteration 60, objective = 0.44695023\n",
"iteration 70, objective = 0.4453482\n",
"iteration 80, objective = 0.444919\n",
"iteration 90, objective = 0.4451945\n",
"iteration 100, objective = 0.44311014\n",
"iteration 110, objective = 0.44101325\n",
"iteration 120, objective = 0.43727913\n",
"iteration 130, objective = 0.43938398\n",
"iteration 140, objective = 0.43788543\n",
"iteration 150, objective = 0.43573555\n",
"iteration 160, objective = 0.4379884\n",
"iteration 170, objective = 0.43852594\n",
"iteration 180, objective = 0.4391472\n",
"iteration 190, objective = 0.4318109\n",
"iteration 200, objective = 0.4389726\n",
"PlayPredictor accuracy: 0.7234723472347235\n"
] ]
} }
], ],
"source": [ "source": [
"model = PlayPredictor()\n",
"model.fit(train, K=6, iters=200)\n",
"\n",
"error = 0\n", "error = 0\n",
"balanced_valid = get_balanced_validation(dataset, valid)\n", "balanced_valid = get_balanced_set(dataset, valid)\n",
"for user, game, review in balanced_valid:\n", "for user, game, review in balanced_valid:\n",
" pred = model.predict(user, game)\n", " pred = model.predict(user, game, threshold=0.5)\n",
" if pred != review[\"played\"]:\n", " if pred != review[\"played\"]:\n",
" error += 1\n", " error += 1\n",
"\n", "\n",
@ -194,7 +299,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 35, "execution_count": 46,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@ -210,7 +315,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 36, "execution_count": 47,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [
@ -287,22 +392,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 40, "execution_count": 48,
"metadata": {},
"outputs": [],
"source": [
"from sklearn.metrics import mean_squared_error\n",
"\n",
"def MSE(y, ypred):\n",
" return mean_squared_error(y, ypred)\n",
"\n",
"model = TimePredictor()\n",
"model.fit(train, l=5.0, iters=200)"
]
},
{
"cell_type": "code",
"execution_count": 41,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{ {
@ -314,6 +404,14 @@
} }
], ],
"source": [ "source": [
"from sklearn.metrics import mean_squared_error\n",
"\n",
"def MSE(y, ypred):\n",
" return mean_squared_error(y, ypred)\n",
"\n",
"model = TimePredictor()\n",
"model.fit(train)\n",
"\n",
"y = []\n", "y = []\n",
"y_pred = []\n", "y_pred = []\n",
"for user, game, review in valid:\n", "for user, game, review in valid:\n",
@ -325,7 +423,7 @@
}, },
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 42, "execution_count": 49,
"metadata": {}, "metadata": {},
"outputs": [], "outputs": [],
"source": [ "source": [

305
assignment1.py
View File

@ -1,15 +1,308 @@
# %% [markdown]
# ### Read Data
# %% # %%
# load data into dataset array
import gzip import gzip
from collections import defaultdict from collections import defaultdict
import json import numpy as np
import tensorflow as tf
# %% # %%
dataset = [] def readJSON(path):
f = gzip.open(path, 'rt', encoding="utf-8")
for l in gzip.open("train.json.gz", 'rt', encoding='utf-8'): f.readline()
for l in f:
d = eval(l) d = eval(l)
dataset.append(d) u = d['userID']
g = d['gameID']
yield u,g,d
print(dataset[0]) dataset = []
for l in readJSON("train.json.gz"):
dataset.append(l)
for user,game,review in dataset:
review["played"] = 1
# %%
# train test split
from sklearn.model_selection import train_test_split
#train, valid = train_test_split(dataset, train_size=165000, random_state=0)
train = dataset[:165000]
valid = dataset[165000:]
# %%
# Get negative labels in vaidation
import random
def get_balanced_set(dataset, s):
all_games = set()
user_played = defaultdict(set)
for user,game,review in dataset:
all_games.add(review["gameID"])
user_played[review["userID"]].add(review["gameID"])
negative = []
for user,game,review in s:
not_played = all_games - user_played[user]
new_game = random.choice(tuple(not_played))
negative.append((user, new_game, {"played": 0}))
return s + negative
# %% [markdown]
# ### Utility Functions
# %%
def writePredictions(infile, outfile, model):
with open(outfile, 'w') as predictions:
for l in open(infile):
if l.startswith("userID"):
predictions.write(l)
continue
u,g = l.strip().split(',')
pred = model.predict(u,g)
_ = predictions.write(u + ',' + g + ',' + str(pred) + '\n')
predictions.close()
# %%
class BPRbatch(tf.keras.Model):
def __init__(self, K, lamb, itemIDs, userIDs):
super(BPRbatch, self).__init__()
# Initialize variables
self.betaI = tf.Variable(tf.random.normal([len(itemIDs)],stddev=0.001))
self.gammaU = tf.Variable(tf.random.normal([len(userIDs),K],stddev=0.001))
self.gammaI = tf.Variable(tf.random.normal([len(itemIDs),K],stddev=0.001))
# Regularization coefficient
self.lamb = lamb
# Prediction for a single instance
def predict(self, u, i):
p = self.betaI[i] + tf.tensordot(self.gammaU[u], self.gammaI[i], 1)
return p
# Regularizer
def reg(self):
return self.lamb * (tf.nn.l2_loss(self.betaI) +\
tf.nn.l2_loss(self.gammaU) +\
tf.nn.l2_loss(self.gammaI))
def score(self, sampleU, sampleI):
u = tf.convert_to_tensor(sampleU, dtype=tf.int32)
i = tf.convert_to_tensor(sampleI, dtype=tf.int32)
beta_i = tf.nn.embedding_lookup(self.betaI, i)
gamma_u = tf.nn.embedding_lookup(self.gammaU, u)
gamma_i = tf.nn.embedding_lookup(self.gammaI, i)
x_ui = beta_i + tf.reduce_sum(tf.multiply(gamma_u, gamma_i), 1)
return x_ui
def call(self, sampleU, sampleI, sampleJ):
x_ui = self.score(sampleU, sampleI)
x_uj = self.score(sampleU, sampleJ)
return -tf.reduce_mean(tf.math.log(tf.math.sigmoid(x_ui - x_uj)))
# %% [markdown]
# ### Play Predictor
# %%
class PlayPredictor:
def __init__(self):
pass
def fit(self, data, threshold=0.6, K=5, iters=100): # data is an array of (user, game, review) tuples
self.topGames = self.getTopGames(threshold)
self.userIDs = {}
self.itemIDs = {}
interactions = []
for u,i,r in data:
if not u in self.userIDs: self.userIDs[u] = len(self.userIDs)
if not i in self.itemIDs: self.itemIDs[i] = len(self.itemIDs)
interactions.append((u,i,r["played"]))
items = list(self.itemIDs.keys())
itemsPerUser = defaultdict(list)
usersPerItem = defaultdict(list)
for u,i,r in interactions:
itemsPerUser[u].append(i)
usersPerItem[i].append(u)
def trainingStepBPR(model, interactions):
Nsamples = 50000
with tf.GradientTape() as tape:
sampleU, sampleI, sampleJ = [], [], []
for _ in range(Nsamples):
u,i,_ = random.choice(interactions) # positive sample
j = random.choice(items) # negative sample
while j in itemsPerUser[u]:
j = random.choice(items)
sampleU.append(self.userIDs[u])
sampleI.append(self.itemIDs[i])
sampleJ.append(self.itemIDs[j])
loss = model(sampleU,sampleI,sampleJ)
loss += model.reg()
gradients = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients((grad, var) for
(grad, var) in zip(gradients, model.trainable_variables)
if grad is not None)
return loss.numpy()
optimizer = tf.keras.optimizers.Adam(0.1)
self.modelBPR = BPRbatch(K, 0.00001, self.itemIDs, self.userIDs)
for i in range(iters):
obj = trainingStepBPR(self.modelBPR, interactions)
if (i % 10 == 9): print("iteration " + str(i+1) + ", objective = " + str(obj))
def predict(self, user, game, threshold=0.5):
if user in self.userIDs and game in self.itemIDs:
pred = self.modelBPR.predict(self.userIDs[user], self.itemIDs[game]).numpy()
return int(pred > threshold)
else:
return int(game in self.topGames)
def getTopGames (self, threshold):
gameCount = defaultdict(int)
totalPlayed = 0
for user,game,_ in readJSON("train.json.gz"):
gameCount[game] += 1
totalPlayed += 1
mostPopular = [(gameCount[x], x) for x in gameCount]
mostPopular.sort()
mostPopular.reverse()
return1 = set()
count = 0
for ic, i in mostPopular:
count += ic
return1.add(i)
if count > totalPlayed * threshold: break
return return1
# %%
model = PlayPredictor()
model.fit(train, K=6, iters=200)
error = 0
balanced_valid = get_balanced_set(dataset, valid)
for user, game, review in balanced_valid:
pred = model.predict(user, game, threshold=0.5)
if pred != review["played"]:
error += 1
print(f"PlayPredictor accuracy: ", 1 - error / len(balanced_valid))
# %%
writePredictions("pairs_Played.csv", "predictions_Played.csv", model)
# %% [markdown]
# ### Time Predictor
# %%
from copy import copy
class TimePredictor:
def __init__(self):
pass
def fit(self, data, l=5.0, iters=200): # data is an array of (user, game, review) tuples
reviewsPerUser = defaultdict(list)
reviewsPerItem = defaultdict(list)
globalAverage = 0
for user, game, review in data:
reviewsPerUser[user].append(review)
reviewsPerItem[game].append(review)
globalAverage += review["hours_transformed"]
globalAverage /= len(data)
betaU = {}
betaI = {}
for u in reviewsPerUser:
reviews = [r["hours_transformed"] for r in reviewsPerUser[u]]
betaU[u] = np.mean(reviews)
for g in reviewsPerItem:
reviews = [r["hours_transformed"] for r in reviewsPerItem[g]]
betaI[g] = np.mean(reviews)
alpha = globalAverage # Could initialize anywhere, this is a guess
for i in range(iters):
newAlpha = 0
for user,game,review in data:
newAlpha += review["hours_transformed"] - (betaU[user] + betaI[game])
alpha = newAlpha / len(data)
for user in reviewsPerUser:
bu = 0
for review in reviewsPerUser[user]:
item = review["gameID"]
bu += review["hours_transformed"] - (alpha + betaI[item])
betaU[user] = bu / (l + len(reviewsPerUser[user]))
for item in reviewsPerItem:
bi = 0
for review in reviewsPerItem[item]:
user = review["userID"]
bi += review["hours_transformed"] - (alpha + betaU[user])
betaI[item] = bi / (l + len(reviewsPerItem[item]))
self.alpha = alpha
self.betaU = betaU
self.betaI = betaI
def predict(self, user, game):
bu = 0
bi = 0
if user in self.betaU:
bu = self.betaU[user]
if game in self.betaI:
bi = self.betaI[game]
return self.alpha + bu + bi
# %%
from sklearn.metrics import mean_squared_error
def MSE(y, ypred):
return mean_squared_error(y, ypred)
model = TimePredictor()
model.fit(train)
y = []
y_pred = []
for user, game, review in valid:
y_pred.append(model.predict(user, game))
y.append(review["hours_transformed"])
print(f"TimePredictor MSE: {MSE(y, y_pred)}")
# %%
writePredictions("pairs_Hours.csv", "predictions_Hours.csv", model)

7050
predictions_Played.csv
View File

File diff suppressed because it is too large Load Diff