diff options
author | 2018-06-03 21:51:32 -0700 | |
---|---|---|
committer | 2018-06-03 21:51:32 -0700 | |
commit | fd9246a308e77c6c27d5bddcc6646525f3ce5e7b (patch) | |
tree | 802568ea9247218a987e91a2eaa3b73004838355 /tensorflow/contrib/tensor_forest | |
parent | 63dafb7f5dbef4da63e095595a49f5d5d7258af9 (diff) |
Switch from tf.contrib.metrics to tf.metrics (#18783)
* Switch from tf.contrib.metrics to tf.metrics
Much of the functions in `tf.contrib.metrics` has been
deprecated in favor of `tf.metrics`. This fix
switches `tf.contrib.metrics` to `tf.metrics`.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
* Change to `tf.metrics` in evaluation_test.py
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
* Fix pylint issue
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Diffstat (limited to 'tensorflow/contrib/tensor_forest')
-rw-r--r-- | tensorflow/contrib/tensor_forest/client/eval_metrics.py | 45 |
1 files changed, 24 insertions, 21 deletions
diff --git a/tensorflow/contrib/tensor_forest/client/eval_metrics.py b/tensorflow/contrib/tensor_forest/client/eval_metrics.py index e893e1d1c8..d8236a0a6f 100644 --- a/tensorflow/contrib/tensor_forest/client/eval_metrics.py +++ b/tensorflow/contrib/tensor_forest/client/eval_metrics.py @@ -21,10 +21,10 @@ import numpy as np from tensorflow.contrib import losses from tensorflow.contrib.learn.python.learn.estimators import prediction_key -from tensorflow.contrib.metrics.python.ops import metric_ops from tensorflow.python.ops import array_ops from tensorflow.python.ops import math_ops +from tensorflow.python.ops import metrics from tensorflow.python.ops import nn INFERENCE_PROB_NAME = prediction_key.PredictionKey.PROBABILITIES @@ -38,12 +38,13 @@ def _top_k_generator(k): targets = math_ops.to_int32(targets) if targets.get_shape().ndims > 1: targets = array_ops.squeeze(targets, axis=[1]) - return metric_ops.streaming_mean(nn.in_top_k(probabilities, targets, k)) + return metrics.mean(nn.in_top_k(probabilities, targets, k)) return _top_k def _accuracy(predictions, targets, weights=None): - return metric_ops.streaming_accuracy(predictions, targets, weights=weights) + return metrics.accuracy( + labels=targets, predictions=predictions, weights=weights) def _r2(probabilities, targets, weights=None): @@ -53,7 +54,7 @@ def _r2(probabilities, targets, weights=None): squares_residuals = math_ops.reduce_sum( math_ops.square(targets - probabilities), 0) score = 1 - math_ops.reduce_sum(squares_residuals / squares_total) - return metric_ops.streaming_mean(score, weights=weights) + return metrics.mean(score, weights=weights) def _squeeze_and_onehot(targets, depth): @@ -62,7 +63,7 @@ def _squeeze_and_onehot(targets, depth): def _sigmoid_entropy(probabilities, targets, weights=None): - return metric_ops.streaming_mean( + return metrics.mean( losses.sigmoid_cross_entropy(probabilities, _squeeze_and_onehot( targets, @@ -71,7 +72,7 @@ def _sigmoid_entropy(probabilities, targets, weights=None): def _softmax_entropy(probabilities, targets, weights=None): - return metric_ops.streaming_mean( + return metrics.mean( losses.sparse_softmax_cross_entropy(probabilities, math_ops.to_int32(targets)), weights=weights) @@ -82,7 +83,7 @@ def _predictions(predictions, unused_targets, **unused_kwargs): def _class_log_loss(probabilities, targets, weights=None): - return metric_ops.streaming_mean( + return metrics.mean( losses.log_loss(probabilities, _squeeze_and_onehot(targets, array_ops.shape(probabilities)[1])), @@ -90,34 +91,36 @@ def _class_log_loss(probabilities, targets, weights=None): def _precision(predictions, targets, weights=None): - return metric_ops.streaming_precision(predictions, targets, weights=weights) + return metrics.precision( + labels=targets, predictions=predictions, weights=weights) def _precision_at_thresholds(predictions, targets, weights=None): - return metric_ops.streaming_precision_at_thresholds( - array_ops.slice(predictions, [0, 1], [-1, 1]), - targets, - np.arange( - 0, 1, 0.01, dtype=np.float32), + return metrics.precision_at_thresholds( + labels=targets, + predictions=array_ops.slice(predictions, [0, 1], [-1, 1]), + thresholds=np.arange(0, 1, 0.01, dtype=np.float32), weights=weights) def _recall(predictions, targets, weights=None): - return metric_ops.streaming_recall(predictions, targets, weights=weights) + return metrics.recall( + labels=targets, predictions=predictions, weights=weights) def _recall_at_thresholds(predictions, targets, weights=None): - return metric_ops.streaming_recall_at_thresholds( - array_ops.slice(predictions, [0, 1], [-1, 1]), - targets, - np.arange( - 0, 1, 0.01, dtype=np.float32), + return metrics.recall_at_thresholds( + labels=targets, + predictions=array_ops.slice(predictions, [0, 1], [-1, 1]), + thresholds=np.arange(0, 1, 0.01, dtype=np.float32), weights=weights) def _auc(probs, targets, weights=None): - return metric_ops.streaming_auc(array_ops.slice(probs, [0, 1], [-1, 1]), - targets, weights=weights) + return metrics.auc( + labels=targets, + predictions=array_ops.slice(probs, [0, 1], [-1, 1]), + weights=weights) _EVAL_METRICS = { |