def fit(self, X_train, y_train):
for i in range(self.epochs):
# update epochs for modified rms updater
self.layers[1].updater.iteration = i
self.layers[2].updater.iteration = i
X_train, y_train = shuffle(X_train, y_train)
for X, y in yield_batches(X_train, y_train, self.batchsize):
self._fit(X, y)
def fit_test(self, X_train, y_train, X_test, y_test, Scaler):
# evaluate before starting training
self._evaluate(X_test, y_test, 0, 0, Scaler)
for i in range(self.epochs):
start = time.time()
# update epochs for modified rms updater
for l in self.layers:
l.epoch = i
X_train, y_train = shuffle(X_train, y_train)
for X, y in yield_batches(X_train, y_train, self.batchsize):
self._fit(X, y)
duration = time.time() - start
self._evaluate(X_test, y_test, i + 1, duration, Scaler)
def _evaluate(self, X_test, y_test, epoch, duration, Scaler):
self.layers[1].b_test = True
result = []
target = []
for X, y in yield_batches(X_test, y_test, self.batchsize):
users = X[:, 0]
movies = X[:, 1]
ratings = y
pred = Scaler.inv(self.predict(users, movies))
result.extend(pred)
target.extend(ratings)
print(np.min(result), np.mean(result), np.max(result))
val_error = RMSE(target, np.clip(result, 1, 5))
self.validation.append(val_error)
print('epoch: {:>2d} | test error: {:>1.5f} | time: {:>5.2f}'.format(
epoch, val_error, duration))
