使用 Tensorflow Serving 和 SavedModel Estimators 获取模型解释

Question

我训练了一个 BoostedTreesClassifier 并且想使用本教程中列出的“定向特征贡献”。 基本上它可以让您“解释”模型的预测并通过使用experimental_predict_with_explanations方法测量每个特征的贡献。 在我训练模型然后调用该方法后效果很好。

但我想用export_saved_model方法导出经过训练的估算器。 当我使用tf.saved_model.load将估算器加载回 Python 环境时，我显然失去了该功能，因为我无法再调用Experiment_predict_with_explanations方法。 加载的模型只有“预测”签名。

最终，我想将此训练有素的估算器与 Tensorflow Serving 结合使用。 我认为它不适用于“预测”SignatureDef。 有没有人试过这个？

Answer 1

带有Tensorflow Serving Trained Estimator可与"Predict" SignatureDef 。

它可以通过使用build_raw_serving_input_receiver_fn而不是build_parsing_serving_input_receiver_fn来实现。

各行代码如下所示：

serving_input_receiver_fn = tf.estimator.export.build_raw_serving_input_receiver_fn(feature_placeholders)

带有 Predict SignatureDef Classification Model完整代码如下所示：

import tensorflow as tf
import iris_data


BATCH_SIZE = 100
STEPS = 1000
Export_Dir = 'Premade_Estimator_Export_Raw' #No need of Version Number

(train_x, train_y), (test_x, test_y) = iris_data.load_data()
type(train_x.values[0][0])

# Feature columns describe how to use the input.
my_feature_columns = []
for key in train_x.keys():
    my_feature_columns.append(tf.feature_column.numeric_column(key=key))

print(my_feature_columns)

columns = [('SepalLength', tf.float32), ('SepalWidth', tf.float32),
           ('PetalLength', tf.float32), ('PetalWidth', tf.float32)]


feature_placeholders = {name: tf.placeholder(dtype, [1], name=name + "_placeholder") for name, dtype in columns}

print(feature_placeholders)

print(type(train_x))

# Build a DNN with 2 hidden layers and 10 nodes in each hidden layer.
classifier = tf.estimator.DNNClassifier(feature_columns=my_feature_columns,    
                                        hidden_units=[10, 10],  # Two hidden layers of 10 nodes each.
                                            n_classes=3) # The model must choose between 3 classes.    

# Train the Model.
classifier.train(input_fn=lambda:iris_data.train_input_fn(train_x, train_y, BATCH_SIZE),steps=STEPS)

eval_result = classifier.evaluate(input_fn=lambda:iris_data.eval_input_fn(test_x, test_y, BATCH_SIZE))

print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result))

# Generate predictions from the model
expected = ['Setosa', 'Versicolor', 'Virginica']
predict_x = {
    'SepalLength': [5.1, 5.9, 6.9],
    'SepalWidth': [3.3, 3.0, 3.1],
    'PetalLength': [1.7, 4.2, 5.4],
    'PetalWidth': [0.5, 1.5, 2.1],
}

predictions = classifier.predict(input_fn=lambda:iris_data.eval_input_fn(features = predict_x, labels = None, 
                                            batch_size=BATCH_SIZE))

template = ('\nPrediction is "{}" ({:.1f}%), expected "{}"')

for pred_dict, expec in zip(predictions, expected):
    class_id = pred_dict['class_ids'][0]
    probability = pred_dict['probabilities'][class_id]

    print(template.format(iris_data.SPECIES[class_id],100 * probability, expec))

# This is the Important Step
serving_input_receiver_fn = tf.estimator.export.build_raw_serving_input_receiver_fn(feature_placeholders)
export_dir = classifier.export_saved_model(Export_Dir, serving_input_receiver_fn)
print('Exported to {}'.format(export_dir))

上述Model的SignatureDef如下图所示：

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:

signature_def['predict']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['PetalLength'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1)
        name: PetalLength_placeholder:0
    inputs['PetalWidth'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1)
        name: PetalWidth_placeholder:0
    inputs['SepalLength'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1)
        name: SepalLength_placeholder:0
    inputs['SepalWidth'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1)
        name: SepalWidth_placeholder:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['all_class_ids'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 3)
        name: dnn/head/predictions/Tile:0
    outputs['all_classes'] tensor_info:
        dtype: DT_STRING
        shape: (-1, 3)
        name: dnn/head/predictions/Tile_1:0
    outputs['class_ids'] tensor_info:
        dtype: DT_INT64
        shape: (-1, 1)
        name: dnn/head/predictions/ExpandDims_2:0
    outputs['classes'] tensor_info:
        dtype: DT_STRING
        shape: (-1, 1)
        name: dnn/head/predictions/str_classes:0
    outputs['logits'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 3)
        name: dnn/logits/BiasAdd:0
    outputs['probabilities'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 3)
        name: dnn/head/predictions/probabilities:0
  Method name is: tensorflow/serving/predict

可以使用以下命令进行Inference ：

sudo docker pull tensorflow/serving

sudo docker run -p 8501:8501 --mount type=bind,source=/usr/local/google/home/Jupyter_Notebooks/TF_Serving/Serving_Made_Easy/Serving_Demystified/Premade_Estimator_Export_Raw,target=/models/Premade_Estimator_Export_Raw -e MODEL_NAME=Premade_Estimator_Export_Raw -t tensorflow/serving &

curl -d '{"signature_name":"predict","instances": [{"SepalLength":[5.1],"SepalWidth":[3.3],"PetalLength":[1.7],"PetalWidth":[0.5]}]}'
-X POST http://localhost:8501/v1/models/Premade_Estimator_Export_Raw:predict

Output如下所示：

{"predictions": [{ "all_classes": ["0", "1", "2"], "probabilities": [0.996251881, 0.00374808488, 3.86118275e-15], "logits": [14.2761269, 8.69337177, -18.9079208], "class_ids": [0], "classes": ["0"], "all_class_ids": [0, 1, 2]}]}