SageMaker Python SDK tensorflow notebooks

Author: mvsusp

sagemaker-python-sdk/tensorflow_abalone_age_predictor_using_keras/abalone.py

@@ -0,0 +1,90 @@
+import numpy as np
+import os
+import tensorflow as tf
+from tensorflow.contrib.keras.python.keras.layers import Dense
+from tensorflow.python.estimator.export.export import build_raw_serving_input_receiver_fn
+from tensorflow.python.estimator.export.export_output import PredictOutput
+
+INPUT_TENSOR_NAME = "inputs"
+SIGNATURE_NAME = "serving_default"
+LEARNING_RATE = 0.001
+
+
+def model_fn(features, labels, mode, params):
+    """Model function for Estimator.
+     # Logic to do the following:
+     # 1. Configure the model via Keras functional api
+     # 2. Define the loss function for training/evaluation using Tensorflow.
+     # 3. Define the training operation/optimizer using Tensorflow operation/optimizer.
+     # 4. Generate predictions as Tensorflow tensors.
+     # 5. Generate necessary evaluation metrics.
+     # 6. Return predictions/loss/train_op/eval_metric_ops in EstimatorSpec object"""
+
+    # 1. Configure the model via Keras functional api
+
+    first_hidden_layer = Dense(10, activation='relu', name='first-layer')(features[INPUT_TENSOR_NAME])
+    second_hidden_layer = Dense(10, activation='relu')(first_hidden_layer)
+    output_layer = Dense(1, activation='linear')(second_hidden_layer)
+
+    predictions = tf.reshape(output_layer, [-1])
+
+    # Provide an estimator spec for `ModeKeys.PREDICT`.
+    if mode == tf.estimator.ModeKeys.PREDICT:
+        return tf.estimator.EstimatorSpec(
+            mode=mode,
+            predictions={"ages": predictions},
+            export_outputs={SIGNATURE_NAME: PredictOutput({"ages": predictions})})
+
+    # 2. Define the loss function for training/evaluation using Tensorflow.
+    loss = tf.losses.mean_squared_error(labels, predictions)
+
+    # 3. Define the training operation/optimizer using Tensorflow operation/optimizer.
+    train_op = tf.contrib.layers.optimize_loss(
+        loss=loss,
+        global_step=tf.contrib.framework.get_global_step(),
+        learning_rate=params["learning_rate"],
+        optimizer="SGD")
+
+    # 4. Generate predictions as Tensorflow tensors.
+    predictions_dict = {"ages": predictions}
+
+    # 5. Generate necessary evaluation metrics.
+    # Calculate root mean squared error as additional eval metric
+    eval_metric_ops = {
+        "rmse": tf.metrics.root_mean_squared_error(
+            tf.cast(labels, tf.float32), predictions)
+    }
+
+    # Provide an estimator spec for `ModeKeys.EVAL` and `ModeKeys.TRAIN` modes.
+    return tf.estimator.EstimatorSpec(
+        mode=mode,
+        loss=loss,
+        train_op=train_op,
+        eval_metric_ops=eval_metric_ops)
+
+
+def serving_input_fn(params):
+    tensor = tf.placeholder(tf.float32, shape=[1, 7])
+    return build_raw_serving_input_receiver_fn({INPUT_TENSOR_NAME: tensor})()
+
+
+params = {"learning_rate": LEARNING_RATE}
+
+
+def train_input_fn(training_dir, params):
+    return _input_fn(training_dir, 'abalone_train.csv')
+
+
+def eval_input_fn(training_dir, params):
+    return _input_fn(training_dir, 'abalone_test.csv')
+
+
+def _input_fn(training_dir, training_filename):
+    training_set = tf.contrib.learn.datasets.base.load_csv_without_header(
+        filename=os.path.join(training_dir, training_filename), target_dtype=np.int, features_dtype=np.float32)
+
+    return tf.estimator.inputs.numpy_input_fn(
+        x={INPUT_TENSOR_NAME: np.array(training_set.data)},
+        y=np.array(training_set.target),
+        num_epochs=None,
+        shuffle=True)()

sagemaker-python-sdk/tensorflow_abalone_age_predictor_using_keras/data/abalone_predict.csv

@@ -0,0 +1,7 @@
+0.17,0.105,0.035,0.034,0.012,0.0085,0.005,4
+0.465,0.39,0.14,0.5555,0.213,0.1075,0.215,15 
+0.365,0.28,0.09,0.196,0.0865,0.036,0.0605,7
+0.525,0.405,0.16,0.658,0.2655,0.1125,0.225,12
+0.585,0.45,0.175,1.1275,0.4925,0.262,0.335,11
+0.44,0.34,0.14,0.482,0.186,0.1085,0.16,9
+0.555,0.445,0.135,0.836,0.336,0.1625,0.275,13