def train(data, user_features=None, item_features=None, use_features=False):
loss_type = "warp" # "bpr"
model = LightFM(learning_rate=0.05, loss=loss_type, max_sampled=100)
if use_features:
model.fit_partial(data,
epochs=20,
user_features=friends_features,
item_features=item_features)
train_precision = precision_at_k(model,
data,
k=10,
user_features=friends_features,
item_features=item_features).mean()
train_auc = auc_score(model,
data,
user_features=friends_features,
item_features=item_features).mean()
print(f'Precision: train {train_precision:.2f}')
print(f'AUC: train {train_auc:.2f}')
else:
model.fit_partial(data, epochs=20)
train_precision = precision_at_k(model, data, k=10).mean()
train_auc = auc_score(model, data).mean()
print(f'Precision: train {train_precision:.2f}')
print(f'AUC: train {train_auc:.2f}')
return model
def evaluate_fm(model_, te_, tr_, items_features=None, users_features=None):
if not tr_.multiply(te_).nnz == 0:
print('train test interaction are not fully disjoin')
# Compute and print the AUC score
train_auc = auc_score(model_,
tr_,
item_features=items_features,
user_features=users_features,
num_threads=NUM_THREADS).mean()
print('Collaborative filtering train AUC: %s' % train_auc)
test_auc = auc_score(model_,
te_,
train_interactions=tr_,
item_features=items_features,
user_features=users_features,
num_threads=NUM_THREADS).mean()
print('Collaborative filtering test AUC: %s' % test_auc)
p_at_k_train = precision_at_k(model_,
tr_,
item_features=items_features,
user_features=users_features,
k=5,
num_threads=NUM_THREADS).mean()
p_at_k_test = precision_at_k(model_,
te_,
train_interactions=tr_,
item_features=items_features,
user_features=users_features,
k=5,
num_threads=NUM_THREADS).mean()
print("Train precision: %.2f" % p_at_k_train)
print("Test precision: %.2f" % p_at_k_test)
def main():
movielens = fetch_movielens()
train = movielens['train']
print(type(train))
print(train.toarray()[:5, :])
test = movielens['test']
print(type(test))
print(test.toarray()[:5, :])
model = LightFM(learning_rate=0.05, loss='bpr')
model.fit(train, epochs=10)
train_precision = precision_at_k(model, train, k=10).mean()
test_precision = precision_at_k(model,
test,
k=10,
train_interactions=train).mean()
train_auc = auc_score(model, train).mean()
test_auc = auc_score(model, test, train_interactions=train).mean()
print(f'train precision: {train_precision}')
print(f'test precision: {test_precision}')
print(f'train auc: {train_auc}')
print(f'test auc: {test_auc}')
print('DONE')
def evaluate(model, train, test, hybrid=False, features=None):
if hybrid:
auc_train = np.mean(auc_score(model, train, item_features=features))
pre_train = np.mean(precision_at_k(model, train, item_features=features))
mrr_train = np.mean(reciprocal_rank(model, train, item_features=features))
auc_test = np.mean(auc_score(model, test, item_features=features))
pre_test = np.mean(precision_at_k(model, test, item_features=features))
mrr_test = np.mean(reciprocal_rank(model, test, item_features=features))
else:
auc_train = np.mean(auc_score(model, train))
pre_train = np.mean(precision_at_k(model, train))
mrr_train = np.mean(reciprocal_rank(model, train))
auc_test = np.mean(auc_score(model, test))
pre_test = np.mean(precision_at_k(model, test))
mrr_test = np.mean(reciprocal_rank(model, test))
res_dict = {'auc_train': auc_train,
'pre_train': pre_train,
'mrr_train': mrr_train,
'auc_test': auc_test,
'pre_test': pre_test,
'mrr_test': mrr_test}
print('The AUC Score is in training/validation: ',
auc_train,' / ', auc_test)
print('The mean precision at k Score in training/validation is: ',
pre_train, ' / ', pre_test)
print('The mean reciprocal rank in training/validation is: ',
mrr_train, ' / ', mrr_test)
print('_________________________________________________________')
return res_dict
def lightfm_model(data, prec_at_k=100, train_split=0.8, epochs=10):
"""
Code to evaluate LightFm model
Data is a scipy sparse matrix
https://arxiv.org/abs/1507.08439
"""
model = LightFM(learning_rate=0.05, loss='logistic')
train, test = random_train_test_split(data,
test_percentage=1 - train_split)
model.fit(train, epochs=epochs) #, num_threads=1)
train_precision = precision_at_k(model, train, k=prec_at_k)
test_precision = precision_at_k(model,
test,
k=prec_at_k,
train_interactions=train)
train_auc = auc_score(model, train)
test_auc = auc_score(model, test, train_interactions=train)
print('Performance of LightFm Model \n')
print(
f'Precision \t Train: {train_precision.mean():.2f} \t Test: {test_precision.mean():.2f}'
)
print(
f'AUC \t\t Train: {train_auc.mean():.2f} \t Test: {test_auc.mean():.2f}'
)
return (train_auc, test_auc, train_precision, test_precision, prec_at_k)
def handler(context):
data = fetch_movielens(min_rating=5.0)
model = LightFM(loss='warp')
epochs = 50
for epoch in range(1, epochs + 1):
print('Epoch: {}'.format(epoch))
model.fit_partial(data['train'], epochs=1, num_threads=1)
train_acc = precision_at_k(model, data['train'], k=5).mean()
test_acc = precision_at_k(model, data['test'], k=5).mean()
print("Train precision: {}".format(train_acc))
print("Test precision: {}".format(test_acc))
statistics = ABEJAStatistics(num_epochs=epochs, epoch=epoch)
statistics.add_stage(ABEJAStatistics.STAGE_TRAIN, float(train_acc),
None)
statistics.add_stage(ABEJAStatistics.STAGE_VALIDATION, float(test_acc),
None)
try:
client.update_statistics(statistics)
except Exception:
pass
np.save(os.path.join(ABEJA_TRAINING_RESULT_DIR, 'model.npy'),
model.__dict__)
def evaluate(self, model, user_items_train):
print("Splitting the data into train/test set...\n")
train, test = cross_validation.random_train_test_split(
user_items_train)
print(train, test)
print("Evaluating methods...\n")
train_recall_10 = recall_at_k(model, train, k=10).mean()
test_recall_10 = recall_at_k(model, test, k=10).mean()
train_recall_20 = recall_at_k(model, train, k=20).mean()
test_recall_20 = recall_at_k(model, test, k=20).mean()
train_precision_10 = precision_at_k(model, train, k=10).mean()
test_precision_10 = precision_at_k(model, test, k=10).mean()
train_precision_20 = precision_at_k(model, train, k=20).mean()
test_precision_20 = precision_at_k(model, test, k=20).mean()
print("Train : Recall@10:{0:.3f}, Recall@20:{1:.3f}".format(
train_recall_10, train_recall_20))
print("Test : Recall@10:{0:.3f}, Recall@20:{1:.3f}".format(
test_recall_10, test_recall_20))
print("Train: Precision@10:{0:.3f}, Precision@20:{1:.3f}".format(
train_precision_10, train_precision_20))
print("Test: Precision@10:{0:.3f}, Precision@20:{1:.3f}".format(
test_precision_10, test_precision_20))
def train():
# import data
data = fetch_movielens(min_rating=5.0)
# https://lyst.github.io/lightfm/docs/datasets.html
# min_rating (float, optional) – Minimum rating to include in the interaction matrix.
# set model super parameter
alpha = 1e-05
epochs = 70
num_components = 32
# https://lyst.github.io/lightfm/docs/lightfm.html
# learning_schedule (string, optional) – one of (‘adagrad’, ‘adadelta’).
warp_model = LightFM(no_components=num_components,
loss='warp',
learning_schedule='adagrad',
max_sampled=100,
user_alpha=alpha,
item_alpha=alpha)
# fitting data
warp_model.fit(data['train'], epochs=30, num_threads=2)
# precision on train, test data
print("Train precision: %.2f" %
precision_at_k(warp_model, data['train'], k=5).mean())
print("Test precision: %.2f" %
precision_at_k(warp_model, data['test'], k=5).mean())
model = warp_model
return model, data
def main():
movielens = fetch_movielens()
train = movielens['train']
test = movielens['test']
print(train.shape)
print(test.shape)
model = LightFM(learning_rate=0.05, loss='bpr')
model.fit(train, epochs=10)
k = 10
train_precision = precision_at_k(model, train, k=k).mean()
test_precision = precision_at_k(model, test, k=k).mean()
print(f'precision_at_{k}(train): {train_precision}')
print(f'precision_at_{k}(test) : {test_precision}')
train_auc = auc_score(model, train).mean()
test_auc = auc_score(model, test).mean()
print(f'auc_score(train): {train_auc}')
print(f'auc_score(test) : {test_auc}')
y_train_preds = model.predict_rank(train)
y_test_preds = model.predict_rank(test)
train_ndcg = ndcg_score(train.toarray(), y_train_preds.toarray())
test_ndcg = ndcg_score(test.toarray(), y_test_preds.toarray())
print(f'ndcg_score(train): {train_ndcg}')
print(f'ndcg_score(test) : {test_ndcg}')
print('DONE')
return 0
def tune():
# import data
data = fetch_movielens(min_rating=5.0)
alpha_ = [1e-05, 1.5e-05, 2e-05]
epochs_ = [50, 60, 70]
num_components_ = [30, 32, 34]
for alpha in alpha_:
for epochs in epochs_:
for num_components in num_components_:
warp_model = LightFM(no_components=num_components,
loss='warp',
learning_schedule='adagrad',
max_sampled=100,
user_alpha=alpha,
item_alpha=alpha)
warp_model.fit(data['train'], epochs=epochs, num_threads=2)
print ('alpha = ', alpha, \
'epochs = ', epochs, \
'num_components = ', num_components )
print("Train precision: %.2f" %
precision_at_k(warp_model, data['train'], k=5).mean())
print("Test precision: %.2f" %
precision_at_k(warp_model, data['test'], k=5).mean())
def test_precision_at_k():
no_users, no_items = (10, 100)
train, test = _generate_data(no_users, no_items)
model = LightFM(loss='bpr')
model.fit_partial(train)
k = 10
# Without omitting train interactions
precision = evaluation.precision_at_k(model, test, k=k)
expected_mean_precision = _precision_at_k(model, test, k)
assert np.allclose(precision.mean(), expected_mean_precision)
assert len(precision) == (test.getnnz(axis=1) > 0).sum()
assert len(evaluation.precision_at_k(model, train,
preserve_rows=True)) == test.shape[0]
# With omitting train interactions
precision = evaluation.precision_at_k(model,
test,
k=k,
train_interactions=train)
expected_mean_precision = _precision_at_k(model, test, k, train=train)
assert np.allclose(precision.mean(), expected_mean_precision)
def evaluate(self, model, train, test, k=10): train_precision = precision_at_k(model, train, k=k).mean() test_precision = precision_at_k(model, test, k=k).mean() train_auc = auc_score(model, train).mean() test_auc = auc_score(model, test).mean() return train_precision, test_precision, train_auc, test_auc
def lightfm_model(data, prec_at_k=10, train_split=0.8):
"""
Code to evaluate LightFm model
Data is a scipy sparse matrix
https://arxiv.org/abs/1507.08439
"""
model = LightFM(learning_rate=0.05, loss='bpr')
train, test = random_train_test_split(data,
test_percentage=1 - train_split)
model.fit(train, epochs=10)
train_precision = precision_at_k(model, train, k=10)
test_precision = precision_at_k(model,
test,
k=10,
train_interactions=train)
train_auc = auc_score(model, train)
test_auc = auc_score(model, test, train_interactions=train)
print('Performance of LightFm Model \n')
print(
f'Precision \t Train: {train_precision.mean():.2f} \t Test: {test_precision.mean():.2f}'
)
print(
f'AUC \t\t Train: {train_auc.mean():.2f} \t Test: {test_auc.mean():.2f}'
)
fig, ax = plt.subplots(2, 2, figsize=(15, 10))
ax[0, 0].hist(train_auc, bins='auto')
ax[0, 0].title.set_text('Distribution of Train AUC score over users')
ax[0, 0].set_ylabel('Count')
ax[0, 0].set_xlabel('AUC Score')
ax[0, 1].hist(test_auc, bins='auto')
ax[0, 1].title.set_text('Distribution of Test AUC score over users')
ax[0, 1].set_ylabel('Count')
ax[0, 1].set_xlabel('AUC Score')
ax[1, 0].hist(train_precision, bins='auto')
ax[1, 0].title.set_text(
f'Distribution of Train Precision @ {prec_at_k} for all users')
ax[1, 0].set_ylabel('Count')
ax[1, 0].set_xlabel(f'Precision @ {prec_at_k}')
ax[1, 1].hist(test_precision, bins='auto')
ax[1, 1].title.set_text(
f'Distribution of Test Precision @ {prec_at_k} for all users')
ax[1, 1].set_ylabel('Count')
ax[1, 1].set_xlabel(f'Precision @ {prec_at_k}')
plt.show()
print('\n')
def collab_filtering():
"""
implements collaborative filtering version
by using only the rating data from movielens dataset
:return:
"""
data = fetch_movielens()
for key, value in data.items():
print(key, type(value), value.shape)
train = data['train']
test = data['test']
print(
'The dataset has %s users and %s items, '
'with %s interactions in the test and %s interactions in the training set.'
% (train.shape[0], train.shape[1], test.getnnz(), train.getnnz()))
model = LightFM(learning_rate=0.05, loss='bpr')
model.fit(train, epochs=50, num_threads=5)
train_precision = precision_at_k(model, train, k=10).mean()
test_precision = precision_at_k(model, test, k=10).mean()
train_recall = recall_at_k(model, test, k=10).mean()
test_recall = recall_at_k(model, test, k=10).mean()
train_auc = auc_score(model, train).mean()
test_auc = auc_score(model, test).mean()
print('Precision: train %.2f, test %.2f.' %
(train_precision, test_precision))
print('Recall: train %.2f, test %.2f.' % (train_recall, test_recall))
print('AUC: train %.2f, test %.2f.' % (train_auc, test_auc))
model = LightFM(learning_rate=0.05, loss='warp')
#resume training from the model's previous state
model.fit_partial(train, epochs=50, num_threads=5)
train_precision = precision_at_k(model, train, k=10).mean()
test_precision = precision_at_k(model, test, k=10).mean()
train_recall = recall_at_k(model, test, k=10).mean()
test_recall = recall_at_k(model, test, k=10).mean()
train_auc = auc_score(model, train).mean()
test_auc = auc_score(model, test).mean()
print("*****************")
print("After re-training")
print('Precision: train %.2f, test %.2f.' %
(train_precision, test_precision))
print('Recall: train %.2f, test %.2f.' % (train_recall, test_recall))
print('AUC: train %.2f, test %.2f.' % (train_auc, test_auc))
#check sample recommendation
sample_recommendation(model, data, [3, 25, 450])
def eval(model, train, val):
# auc
print("Train auc: %.2f" % auc_score(model, train).mean())
print("Val auc: %.2f" % auc_score(model, val).mean())
# precision_at_k
print("Train precision: %.2f" % precision_at_k(model, train, k=5).mean())
print("Val precision: %.2f" % precision_at_k(model, val, k=5).mean())
# recall_at_k
print("Train recall: %.2f" % precision_at_k(model, train, k=5).mean())
print("Val recall: %.2f" % precision_at_k(model, val, k=5).mean())
def patk_learning_curve(model,
train,
test,
iterarray,
user_features=None,
item_features=None,
k=5,
**fit_params):
old_epoch = 0
train_patk = []
test_patk = []
warp_duration = []
# bpr_duration = []
train_warp_auc = []
test_warp_auc = []
# bpr_auc = []
headers = ['Epoch', 'train p@5', 'train_auc', 'test p@5', 'test_auc']
print_log(headers, header=True)
for epoch in iterarray:
more = epoch - old_epoch
start = time.time()
model.fit_partial(train,
user_features=user_features,
epochs=more,
item_features=item_features,
**fit_params)
warp_duration.append(time.time() - start)
train_warp_auc.append(
auc_score(model, train, item_features=item_features).mean())
test_warp_auc.append(
auc_score(model,
test,
item_features=item_features,
train_interactions=train).mean())
this_test = precision_at_k(model,
test,
train_interactions=train,
item_features=item_features,
k=k)
this_train = precision_at_k(model,
train,
train_interactions=None,
item_features=item_features,
k=k)
train_patk.append(np.mean(this_train))
test_patk.append(np.mean(this_test))
row = [
epoch, train_patk[-1], train_warp_auc[-1], test_patk[-1],
test_warp_auc[-1]
]
print_log(row)
return model, train_patk, test_patk, warp_duration, train_warp_auc, test_warp_auc
def evaluate_model(model, train, test, item_fetures=None, user_features=None, num_threads=1):
train_precision = precision_at_k(model, train, k=10, user_features=user_features, item_features=item_fetures, num_threads=num_threads).mean()
test_precision = precision_at_k(model, test, train_interactions=train, k=10, user_features=user_features, item_features=item_fetures, num_threads=num_threads).mean()
train_auc = auc_score(model, train, user_features=user_features, item_features=item_fetures, num_threads=num_threads).mean()
test_auc = auc_score(model, test, train_interactions=train, user_features=user_features, item_features=item_fetures, num_threads=num_threads).mean()
print('Precision: train %.2f, test %.2f.' % (train_precision, test_precision))
print('AUC: train %.2f, test %.2f.' % (train_auc, test_auc))
return train_precision, test_precision, train_auc, test_auc
def measure_accuracies(model, data):
print("\nMeasuring accuracies of the model...")
# evaluate the precision@k metric
training_precision = precision_at_k(model, data["train"], k=PRECISION_K).mean()
test_precision = precision_at_k(model, data["test"], k=PRECISION_K).mean()
# evaluate the AUROC metric
training_auc = auc_score(model, data["train"]).mean()
test_auc = auc_score(model, data["test"]).mean()
# print them out
print("Precision@k: training %.2f, test %.2f" % (training_precision, test_precision))
print("AUC: training %.2f, test %.2f" % (training_auc, test_auc))
def _get_metrics(model, train_set, test_set):
train_set = train_set.tocsr()
test_set = test_set.tocsr()
train_set.data[train_set.data < 0] = 0.0
test_set.data[test_set.data < 0] = 0.0
train_set.eliminate_zeros()
test_set.eliminate_zeros()
return (precision_at_k(model, train_set).mean(),
precision_at_k(model, test_set).mean(),
auc_score(model, train_set).mean(),
auc_score(model, test_set).mean())
def evaluate_model(model, metric, test, train):
"""
Evaluate trained model on the test set, using one of the three available accuracy metrics
AUC: the probability that a randomly chosen positive example has a higher score than a randomly chosen
negative example.
Precision: the fraction of known positives in the first k positions of the ranked list of results.
Recall: the number of positive items in the first k positions of the ranked list of results divided by the
number of positive items in the test period.
:param model:(LightFM, required) - model to be evaluated
:param metric:(string, required) - accuracy metric to be used, one of ['auc', 'precision', 'recall']
:param test:(COO matrix, required) - known positives used to test the model
:param train:(COO matrix, required) - training set; these interactions will be omitted from the score
calculations to avoid re-recommending known positives.
:return: test_score (float) - score computed on the test set
"""
try:
# make sure the metric is correct
assert metric in ['auc', 'precision', 'recall']
if metric == 'auc':
test_score = auc_score(model, test, train).mean()
elif metric == 'precision':
test_score = precision_at_k(model, test, train, k=5).mean()
else:
test_score = recall_at_k(model, test, train, k=5).mean()
return test_score
except AssertionError:
print('The metric provided is not correct or available!')
def best_reccomendation():
#define variables
best = 0.0
best_model = ''
for model in models:
score = 0.0
pak_score = evaluation.precision_at_k(model, data2['test'])
score += np.mean(pak_score)
rak_score = evaluation.recall_at_k(model, data2['test'])
score += np.mean(rak_score)
auc_score = evaluation.auc_score(model, data2['test'])
score += np.mean(auc_score)
rr_score = evaluation.reciprocal_rank(model, data2['test'])
score += np.mean(rr_score)
print(score)
if score >= best:
best = score
best_model = model
return best_model
def test_LightFM_model(model,
test_interactions,
train_interactions,
user_features,
movie_features,
k=5):
test_precision = precision_at_k(model,
test_interactions,
train_interactions,
k=k,
user_features=user_features,
item_features=movie_features,
num_threads=2).mean()
test_recall = recall_at_k(model,
test_interactions,
train_interactions,
k=k,
user_features=user_features,
item_features=movie_features,
num_threads=2).mean()
test_auc = auc_score(model,
test_interactions,
train_interactions,
user_features=user_features,
item_features=movie_features,
num_threads=2).mean()
print('Model')
print('Precision at k=', str(k), ': ', round(test_precision, 3), sep='')
print('Recall at k=', str(k) + ': ', round(test_recall, 3), sep='')
print('AUC: ', round(test_auc, 3), sep='')
return ({
'precision': round(test_precision, 3),
'recall': round(test_recall, 3),
'auc': round(test_auc, 3)
})
def objective(self, params):
epochs, learning_rate, no_components, alpha = params
user_alpha = alpha
item_alpha = alpha
model = LightFM(loss='warp',
random_state=2016,
learning_rate=learning_rate,
no_components=no_components,
user_alpha=user_alpha,
item_alpha=item_alpha)
model.fit(self.train_test_user[0],
epochs=epochs,
num_threads=4,
verbose=True)
patks = evaluation.precision_at_k(model,
self.train_test_user[1],
train_interactions=None,
k=5,
num_threads=4)
print("running hyperparmeter.." +
datetime.datetime.now().strftime("%Y-%M-%d %H:%m"))
mapatk = np.mean(patks)
# Make negative because we want to _minimize_ objective
out = -mapatk
# Handle some weird numerical shit going on
if np.abs(out + 1) < 0.01 or out < -1.0:
return 0.0
else:
return out
def run_BPR(split, user_item, prec_rec_at):
# train_sparse = scipy.sparse.csr_matrix(train_splt.values)
# test_sparse = scipy.sparse.csr_matrix(test_split.values)
train = user_item
test = user_item * 0
for index, row in split.iterrows():
test['u' + str(row['userId'])]['m' +
str(row['movieId'])] = row['rating']
train['u' + str(row['userId'])]['m' + str(row['movieId'])] = 0
train_sparse = scipy.sparse.csr_matrix(train.values)
test_sparse = scipy.sparse.csr_matrix(test.values)
model = LightFM(loss='bpr')
model.fit(train_sparse, epochs=30, num_threads=4)
# print("####################################################################################")
# print("Test precision: %.2f" % precision_at_k(model, test_sparse, k=prec_rec_at).mean())
# print("Test recall: %.2f" % recall_at_k(model, test_sparse, k=prec_rec_at).mean())
# print("####################################################################################")
precision = precision_at_k(model, test_sparse, k=prec_rec_at).mean()
recall = recall_at_k(model, test_sparse, k=prec_rec_at).mean()
return precision, recall
def objective(params):
no_components, learning_rate = params
global model
model = LightFM(no_components=no_components,
learning_schedule='adagrad',
loss='warp',
learning_rate=learning_rate,
random_state=0)
model.fit(interactions=train,
item_features=item_features,
sample_weight=train_weights,
epochs=3,
verbose=False)
test_precision = precision_at_k(model, test, k=5, item_features=item_features).mean()
print("no_comp: {}, lrn_rate: {:.5f}, precision: {:.5f}".format(
no_components, learning_rate, test_precision))
# test_auc = auc_score(model, test, item_features=item_features).mean()
output = -test_precision
if np.abs(output+1) < 0.01 or output < -1.0:
output = 0.0
return output
def train (self, interactions, test_percentage=0.25,
n_components=30, learning_rate = 0.5, loss='warp', model_k=15, n_jobs = 4,
epoch=30, evaluate_k = 50):
from lightfm.evaluation import precision_at_k
from lightfm.evaluation import recall_at_k
from lightfm.cross_validation import random_train_test_split
train, test = random_train_test_split(interactions, test_percentage=test_percentage, random_state=None)
mf_model = self.runMF(interactions = train,
n_components = n_components,
learning_rate = learning_rate,
loss = loss,
k = model_k,
epoch = epoch,
n_jobs = n_jobs)
precise = precision_at_k(mf_model, test_interactions = test, k = evaluate_k)
recall = recall_at_k(mf_model, test_interactions = test, k = evaluate_k)
precise_test = precise.mean()
recall_test = recall.mean()
return mf_model, precise_test, recall_test
def score(self, X, y, **kwargs):
test = to_sparse_matrix(X[:, 0], X[:, 1], y, self.shape)
return precision_at_k(self.model,
test,
train_interactions=self.train,
k=10).mean()
def resultados_colaborativo(self):
"""
Método resultados_colaboraivo. Obtiene los resultados del modelo colaborativo.
Este método solo se utiliza en la interfaz de texto.
"""
global train, test, modelo
# Se obtienen los resultados
precision = precision_at_k(modelo,
test,
train_interactions=train,
k=10,
num_threads=self.CPU_THREADS).mean()
auc = auc_score(modelo,
test,
train_interactions=train,
num_threads=self.CPU_THREADS).mean()
recall = recall_at_k(modelo,
test,
train_interactions=train,
k=10,
num_threads=self.CPU_THREADS).mean()
reciprocal = reciprocal_rank(modelo,
test,
train_interactions=train,
num_threads=self.CPU_THREADS).mean()
# Se imprimen los resultados
imprimir_resultados_clasico(precision, auc, recall, reciprocal)
def main():
# find optimal learning rate and rank
bounds = [(10**-4, 1.0, 'log-uniform'),(10**-6, 10**-1, 'log-uniform')]
opt_params = forest_minimize(optimal, bounds, verbose=True)
opt_lr,opt_rank=opt_params.x[0],opt_params.x[1]
# times and precisions for 3 data sets with
data_1_train,_,data_1_test = read_train_data("0.1_percent")
data_2_train,_, data_2_test = read_train_data("0.5_percent")
data_3_train,_, data_3_test = read_train_data("1_percent")
dataset= [[data_1_train,data_1_test],[data_2_train,data_2_test],[data_3_train,data_3_test]]
times =[]
precisions=[]
for data in dataset:
train,test = sparse_train_test(data[0],data[1])
start = time.time()
model = LightFM(loss='warp', learning_rate=opt_lr,no_components=opt_rank)
model.fit(train, epochs=10, verbose=True)
precision = precision_at_k(model, test, k=50).mean()
sec = (time.time() - start)/60
print("TIME:",sec,"PREC:", precision)
times.append(sec)
precisions.append(np.mean(precision))
print("times:", times, "precisions", precisions)
def experiment(type_exp):
global train, test
model = LightFM(learning_rate=0.05, loss=type_exp)
model.fit(train, epochs=10)
train_precision = precision_at_k(model, train, k=10).mean()
test_precision = precision_at_k(model, test, k=10).mean()
train_auc = auc_score(model, train).mean()
test_auc = auc_score(model, test).mean()
output = open("mid100mostRaiting5.txt", 'w')
output.write("La presicion del training set es: " + str(train_precision) + "\n")
output.write("La presicion del test set es: " + str(test_precision) + "\n")
output.write("El auc del training set es: " + str(train_auc) + "\n")
output.write("El auc del test set es: " + str(test_auc) + "\n")
output.close()
