python - Tensorflow - Extract metric keys and/or dict from tf.contrib.learn.DNNLinearCombinedRegressor -


i having problem getting metrics pre-built estimator models.this code extract metrics model:

v_metrics = {     "accuracy":         tf.contrib.learn.metricspec(             metric_fn=tf.contrib.metrics.streaming_accuracy,             prediction_key = tf.contrib.learn.predictionkey.scores),     "precision":         tf.contrib.learn.metricspec(             metric_fn=tf.contrib.metrics.streaming_precision,             prediction_key = tf.contrib.learn.predictionkey.scores),     "recall":         tf.contrib.learn.metricspec(             metric_fn=tf.contrib.metrics.streaming_recall,             prediction_key = tf.contrib.learn.predictionkey.scores) }  v_monitor = tf.contrib.learn.monitors.validationmonitor(                 input_fn=lambda: input_fn(df_test),                 eval_steps = 1,                 every_n_steps=10,                 metrics=v_metrics )   m = build_estimator(model_dir, model_type) m.fit(input_fn=lambda: input_fn(df_train), steps=train_steps, monitors=[v_monitor]) results = m.evaluate(input_fn=lambda: input_fn(df_test), steps=1) 

however, when run gives me:

info:tensorflow:saving dict global step 28: accuracy = 0.0, global_step = 28, loss = 24.7031, precision = 1.0, recall = 1.0 

i looked api , estimator.py , metric_spec.py scripts, did find way extract tensors responsible accuracy, precision, , recall. looked @ other answers, them solutions custom estimators. checked results on tensorboard , still 0s , 1s.

note when used exact same method tf.contrib.learn.dnnlinearcombinedregressor prediction_key = tf.contrib.learn.predictionkey.classes works properly.

i not sure doing wrong combined regressor. know proper prediction keys (or doing wrong), when run model, gives me proper metric results.


Comments

Popular posts from this blog

PHP and MySQL WP -

android - InAppBilling registering BroadcastReceiver in AndroidManifest -

nginx - phpPgAdmin - log in works but I have to login again after clicking on any links -