Add Iris TensorFlow example (#208)

Author: 1vn

docs/apis/packaging-models.md

@@ -5,7 +5,7 @@
 Zip the exported estimator output in your checkpoint directory:
 
 ```text
-$ ls export/estimator
+$ ls export/estimator/1560263597/
 saved_model.pb  variables/
 
 $ zip -r model.zip export/estimator

examples/iris/tensorflow/model.py

@@ -0,0 +1,94 @@
+# sources copied/modified from https://github.com/tensorflow/models/blob/master/samples/core/get_started/
+
+import tensorflow as tf
+from sklearn.datasets import load_iris
+from sklearn.model_selection import train_test_split
+import shutil
+import os
+
+EXPORT_DIR = "iris_tf_export"
+
+
+def input_fn(features, labels, batch_size, mode):
+    """An input function for training"""
+    dataset = tf.data.Dataset.from_tensor_slices((features, labels))
+    if mode == tf.estimator.ModeKeys.TRAIN:
+        dataset = dataset.shuffle(1000).repeat()
+    dataset = dataset.batch(batch_size)
+    dataset_it = dataset.make_one_shot_iterator()
+    irises, labels = dataset_it.get_next()
+    return {"irises": irises}, labels
+
+
+def json_serving_input_fn():
+    inputs = tf.placeholder(shape=[4], dtype=tf.float64)
+    features = {"irises": tf.expand_dims(inputs, 0)}
+    return tf.estimator.export.ServingInputReceiver(features=features, receiver_tensors=inputs)
+
+
+def my_model(features, labels, mode, params):
+    """DNN with three hidden layers and learning_rate=0.1."""
+    net = features["irises"]
+    for units in params["hidden_units"]:
+        net = tf.layers.dense(net, units=units, activation=tf.nn.relu)
+
+    logits = tf.layers.dense(net, params["n_classes"], activation=None)
+
+    predicted_classes = tf.argmax(logits, 1)
+    if mode == tf.estimator.ModeKeys.PREDICT:
+        predictions = {
+            "class_ids": predicted_classes[:, tf.newaxis],
+            "probabilities": tf.nn.softmax(logits),
+            "logits": logits,
+        }
+        return tf.estimator.EstimatorSpec(
+            mode=mode,
+            predictions=predictions,
+            export_outputs={
+                "predict": tf.estimator.export.PredictOutput(
+                    {
+                        "class_ids": predicted_classes[:, tf.newaxis],
+                        "probabilities": tf.nn.softmax(logits),
+                    }
+                )
+            },
+        )
+
+    loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)
+
+    accuracy = tf.metrics.accuracy(labels=labels, predictions=predicted_classes, name="acc_op")
+    metrics = {"accuracy": accuracy}
+    tf.summary.scalar("accuracy", accuracy[1])
+
+    if mode == tf.estimator.ModeKeys.EVAL:
+        return tf.estimator.EstimatorSpec(mode, loss=loss, eval_metric_ops=metrics)
+
+    optimizer = tf.train.AdagradOptimizer(learning_rate=0.1)
+    train_op = optimizer.minimize(loss, global_step=tf.train.get_global_step())
+    return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op)
+
+
+iris = load_iris()
+X, y = iris.data, iris.target
+X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.8, random_state=42)
+
+classifier = tf.estimator.Estimator(
+    model_fn=my_model, model_dir=EXPORT_DIR, params={"hidden_units": [10, 10], "n_classes": 3}
+)
+
+
+train_input_fn = lambda: input_fn(X_train, y_train, 100, tf.estimator.ModeKeys.TRAIN)
+eval_input_fn = lambda: input_fn(X_test, y_test, 100, tf.estimator.ModeKeys.EVAL)
+serving_input_fn = lambda: json_serving_input_fn()
+exporter = tf.estimator.FinalExporter("estimator", serving_input_fn, as_text=False)
+train_spec = tf.estimator.TrainSpec(train_input_fn, max_steps=1000)
+eval_spec = tf.estimator.EvalSpec(eval_input_fn, exporters=[exporter], name="estimator-eval")
+
+tf.estimator.train_and_evaluate(classifier, train_spec, eval_spec)
+
+# zip the estimator export dir (the exported path looks like iris_tf_export/export/estimator/1562353043/)
+estimator_dir = EXPORT_DIR + "/export/estimator"
+shutil.make_archive("tensorflow", "zip", os.path.join(estimator_dir))
+
+# clean up
+shutil.rmtree(EXPORT_DIR)

examples/iris/tensorflow/requirements.txt

@@ -0,0 +1,2 @@
+tensorflow
+sklearn

pkg/workloads/tf_api/api.py

@@ -391,6 +391,30 @@ def predict(deployment_name, api_name):
     return jsonify(response)
 
 
+def validate_model_dir(model_dir):
+    """
+    validates that model_dir has the expected directory tree.
+
+    For example (your TF serving version number may be different):
+
+    1562353043/
+        saved_model.pb
+        variables/
+            variables.data-00000-of-00001
+            variables.index
+    """
+    version = os.listdir(model_dir)[0]
+    if not version.isdigit():
+        raise UserException(
+            "No versions of servable default found under base path in model_dir. See docs.cortex.dev for how to properly package your TensorFlow model"
+        )
+
+    if "saved_model.pb" not in os.listdir(os.path.join(model_dir, version)):
+        raise UserException(
+            'Expected packaged model to have a "saved_model.pb" file. See docs.cortex.dev for how to properly package your TensorFlow model'
+        )
+
+
 def start(args):
     ctx = Context(s3_path=args.context, cache_dir=args.cache_dir, workload_id=args.workload_id)
 
@@ -406,8 +430,7 @@ def start(args):
             package.install_packages(ctx.python_packages, ctx.storage)
         if not os.path.isdir(args.model_dir):
             ctx.storage.download_and_unzip_external(api["model"], args.model_dir)
-
-    if util.is_resource_ref(api["model"]):
+    else:
         package.install_packages(ctx.python_packages, ctx.storage)
         model_name = util.get_resource_ref(api["model"])
         model = ctx.models[model_name]
@@ -446,6 +469,12 @@ def start(args):
                 model["input"]["target_vocab"], None, False
             )
 
+    try:
+        validate_model_dir(args.model_dir)
+    except Exception as e:
+        logger.exception(e)
+        sys.exit(1)
+
     channel = grpc.insecure_channel("localhost:" + str(args.tf_serve_port))
     local_cache["stub"] = prediction_service_pb2_grpc.PredictionServiceStub(channel)