def eval_metrics_fn():
return {
"accuracy": lambda labels, predictions: accuracy(
tf.cast(tf.squeeze(tf.round(predictions)), tf.int32),
tf.cast(labels, tf.int32),
)
}
def model_fn(features, labels, mode, params):
scores = predict_scores(features)
if mode == ModeKeys.INFER:
return EstimatorSpec(mode, predictions=scores)
positive_scores = lookup_positives(scores, labels['click_position'])
logits = create_diffs(positive_scores, scores)
lbls = create_label(labels['click_position'])
ele_loss = elementwise_loss(lbls, logits, labels['normal_mask']) * lbls
loss = reduce_sum(ele_loss)
true_lbl = true_label(features, labels)
if mode == ModeKeys.EVAL:
return EstimatorSpec(mode,
loss=loss,
eval_metric_ops={
'acc':
mean(
accuracy(
argmax(noise_label(labels), axis=1),
argmax(to_one_hot(scores), axis=1)))
})
else:
optimizer = AdamOptimizer(learning_rate=params['learning_rate'])
train_op = optimizer.minimize(loss, global_step=get_global_step())
return EstimatorSpec(mode, loss=loss, train_op=train_op)