args.dataset,
args.dataset_info_path,
)
fold_id = (args.fold_id-1)%data.num_folds()
data = data.get_data_folds()[fold_id]
start = time.time()
data.read_data()
print('Time past for reading data: %d seconds' % (time.time() - start))
max_ranking_size = np.min((cutoff, data.max_query_size()))
model_params = {'hidden units': [32, 32],
'learning_rate': 0.01,}
model = nn.init_model(model_params)
optimizer = tf.keras.optimizers.SGD(learning_rate=model_params['learning_rate'])
results = []
metric_weights = 1./np.log2(np.arange(max_ranking_size) + 2)
train_labels = 2**data.train.label_vector-1
vali_labels = 2**data.validation.label_vector-1
test_labels = 2**data.test.label_vector-1
train_ndoc_feat = np.zeros(data.train.num_docs())
vali_ndoc_feat = np.zeros(data.validation.num_docs())
test_ndoc_feat = np.zeros(data.test.num_docs())
for qid in range(data.train.num_queries()):
q_feat = data.train.query_values_from_vector(qid, train_ndoc_feat)
q_feat[:] = q_feat.shape[0]
print('Time past for reading data: %d seconds' % (time.time() - start))
max_ranking_size = np.min((cutoff, data.max_query_size()))
click_model = clk.get_click_model(click_model_name)
alpha, beta = click_model(np.arange(max_ranking_size))
true_train_doc_weights = data.train.label_vector * 0.25
true_vali_doc_weights = data.validation.label_vector * 0.25
true_test_doc_weights = data.test.label_vector * 0.25
model_params = {
'hidden units': [32, 32],
}
model = nn.init_model(model_params)
logging_model = nn.init_model(model_params)
# optimizer = tf.keras.optimizers.Adam()
optimizer = tf.keras.optimizers.SGD(learning_rate=0.01)
model.build(input_shape=data.train.feature_matrix.shape)
logging_model.build(input_shape=data.train.feature_matrix.shape)
if args.pretrained_model:
model.load_weights(args.pretrained_model)
logging_model.load_weights(args.pretrained_model)
init_weights = model.get_weights()
init_opt_weights = optimizer.get_weights()
n_sampled = 0
train_clicks = np.zeros((data.train.num_docs(), max_ranking_size),
dtype=np.int64)