Убыток уменьшается, но модель не покрывает - PullRequest
Новая
       18

Убыток уменьшается, но модель не покрывает

1 голос
/ 19 июня 2019

Я пытаюсь реализовать просто модель прогнозирования ctr, используя тензор потока. Тем не менее, потеря тренировки уменьшается, но точность остается той же. (На самом деле он точно равен скорости положительных или отрицательных образцов, кажется, что модель прогнозирует 0 или 1 для всех образцов). Что я должен сделать, чтобы решить эту проблему? 

def parse_exmp(serial_exmp):
    click = fc.numeric_column("click", dtype=tf.float32)
    fea_columns = [click]
    fea_columns += my_feature_columns
    feature_spec=tf.feature_column.make_parse_example_spec(fea_columns)
    feats = tf.parse_single_example(serial_exmp, features=feature_spec)
    click = feats.pop('click')
    return feats, {'ctr': tf.to_float(click)}


def train_input_fn(filenames, batch_size, shuffle_buffer_size):
    dataset = tf.data.TFRecordDataset(filenames)
    dataset = dataset.map(parse_exmp, num_parallel_calls=5)
    dataset = dataset.repeat().batch(batch_size).prefetch(1000)
    return dataset


def eval_input_fn(filename, batch_size):
    dataset = tf.data.TFRecordDataset(filename)
    dataset = dataset.map(parse_exmp, num_parallel_calls=5)
    # Shuffle, repeat, and batch the examples.
    dataset = dataset.batch(batch_size)
    # Return the read end of the pipeline.
    return dataset


def build_model(features, mode, params):
    net = fc.input_layer(features, params['feature_columns'])
    # Build the hidden layers, sized according to the 'hidden_units' param.

    for units in params['hidden_units']:
    net = tf.layers.dense(net, units=units, kernel_regularizer=None, 
                              bias_regularizer=None, activation=tf.nn.relu)
    if 'dropout_rate' in params and params['dropout_rate'] > 0.0:
        net = tf.layers.dropout(net, params['dropout_rate'], training=(mode == tf.estimator.ModeKeys.TRAIN))

    # Compute logits
    logits = tf.layers.dense(net, 1, activation=None)
    return logits

def my_model(features, labels, mode, params):
    ctr_logits = build_model(features, mode, params)
    ctr_predictions = tf.sigmoid(ctr_logits, name="CTR")

    if mode == tf.estimator.ModeKeys.PREDICT:
        predictions = ctr_predictions
        export_outputs = {
            'prediction': tf.estimator.export.PredictOutput(predictions)
        }
        return tf.estimator.EstimatorSpec(mode, predictions=predictions, export_outputs=export_outputs)

    loss = tf.reduce_sum(tf.nn.sigmoid_cross_entropy_with_logits(labels=labels['ctr'], logits=ctr_logits), name="loss")

    if mode == tf.estimator.ModeKeys.TRAIN:
        optimizer = tf.train.AdagradOptimizer(learning_rate=params['learning_rate'])
        train_op = optimizer.minimize(loss, global_step=tf.train.get_global_step())
        return tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)

    ctr_accuracy = tf.metrics.accuracy(labels=labels['ctr'], predictions=tf.to_float(tf.greater_equal(ctr_predictions, 0.5)))
    ctr_auc = tf.metrics.auc(labels['ctr'], ctr_predictions)

    metrics = {'ctr_accuracy': ctr_accuracy, 'ctr_auc': ctr_auc}
    tf.summary.scalar('ctr_accuracy', ctr_accuracy[1])
    tf.summary.scalar('ctr_auc', ctr_auc[1])

    if mode == tf.estimator.ModeKeys.EVAL:
        return tf.estimator.EstimatorSpec(mode, loss=loss, eval_metric_ops=metrics)


create_feature_columns()
classifier = tf.estimator.Estimator(
    model_fn=my_model,
    params={
            'feature_columns': my_feature_columns,
            'hidden_units': hidden_units.split(','),
            'learning_rate': learning_rate,
            'dropout_rate': dropout_rate
    },
    config=tf.estimator.RunConfig(model_dir=model_dir, save_checkpoints_steps=save_checkpoints_steps)
)
batch_size = batch_size


train_spec = tf.estimator.TrainSpec(
    input_fn=lambda: train_input_fn(train_files, batch_size, shuffle_buffer_size),
    max_steps=train_steps
)
input_fn_for_eval = lambda: eval_input_fn(eval_files, batch_size)
eval_spec = tf.estimator.EvalSpec(input_fn=input_fn_for_eval, throttle_secs = 60, steps=None)

Вот лог: 

INFO:tensorflow:Running training and evaluation locally (non-distributed).
INFO:tensorflow:Start train and evaluate loop. The evaluate will happen after 60 secs (eval_spec.throttle_secs) or training is finished.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Saving checkpoints for 1 into ./model_dir/model.ckpt.
INFO:tensorflow:loss = 8575.396, step = 1
INFO:tensorflow:Saving checkpoints for 45 into ./model_dir/model.ckpt.
INFO:tensorflow:Loss for final step: 92.291855.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2019-06-10-07:05:03
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from ./model_dir/model.ckpt-45
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Finished evaluation at 2019-06-10-07:05:17
INFO:tensorflow:Saving dict for global step 45: ctr_accuracy = 0.436, ctr_auc = 0.5, global_step = 45, loss = 6467.209
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from ./model_dir/model.ckpt-45
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Saving checkpoints for 46 into ./model_dir/model.ckpt.
INFO:tensorflow:loss = 1430.6135, step = 46
INFO:tensorflow:Saving checkpoints for 87 into ./model_dir/model.ckpt.
INFO:tensorflow:Loss for final step: 469.1917.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2019-06-10-07:06:44
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from ./model_dir/model.ckpt-87
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Finished evaluation at 2019-06-10-07:06:58
INFO:tensorflow:Saving dict for global step 87: ctr_accuracy = 0.436, ctr_auc = 0.5, global_step = 87, loss = 3759.593
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from ./model_dir/model.ckpt-87
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Saving checkpoints for 88 into ./model_dir/model.ckpt.
INFO:tensorflow:loss = 938.0892, step = 88
INFO:tensorflow:Saving checkpoints for 131 into ./model_dir/model.ckpt.
INFO:tensorflow:Loss for final step: 3.707542e-12.
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Done calling model_fn.
INFO:tensorflow:Starting evaluation at 2019-06-10-07:08:26
INFO:tensorflow:Graph was finalized.
INFO:tensorflow:Restoring parameters from ./model_dir/model.ckpt-131
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Finished evaluation at 2019-06-10-07:08:39
INFO:tensorflow:Saving dict for global step 131: ctr_accuracy = 0.436, ctr_auc = 0.5, global_step = 131, loss = 2633.0947
...