Update iris example, fix estimator bugs

Author: deliahu

examples/iris/app.yaml

@@ -1,2 +1,28 @@
 - kind: app
   name: iris
+
+- kind: environment
+  name: dev
+  data:
+    type: csv
+    path: s3a://cortex-examples/iris.csv
+    schema: [@sepal_length, @sepal_width, @petal_length, @petal_width, @class]
+
+- kind: model
+  name: dnn
+  estimator: cortex.dnn_classifier
+  target_column: @class
+  input:
+    numeric_columns: [@sepal_length, @sepal_width, @petal_length, @petal_width]
+    target_vocab: ['Iris-setosa', 'Iris-versicolor', 'Iris-virginica']
+  hparams:
+    hidden_units: [4, 2]
+  training:
+    batch_size: 10
+    num_steps: 1000
+
+- kind: api
+  name: iris-type
+  model: @dnn
+  compute:
+    replicas: 1

examples/iris/resources/aggregates.yaml

@@ -59,7 +59,7 @@ def create_estimator(run_config, model_config):
             )
         )
 
-    return tf.estimator.BoostedTreesClassifier(
+    return tf.estimator.BoostedTreesRegressor(
         feature_columns=feature_columns,
         n_batches_per_layer=model_config["hparams"]["batches_per_layer"],
         weight_column=model_config["input"].get("weight_column", None),

examples/iris/resources/apis.yaml

@@ -5,7 +5,7 @@ def create_estimator(run_config, model_config):
     dnn_feature_columns = []
 
     for col_name in model_config["input"]["dnn_columns"]["numeric_columns"]:
-        feature_columns.append(tf.feature_column.numeric_column(col_name))
+        dnn_feature_columns.append(tf.feature_column.numeric_column(col_name))
 
     for col_info in model_config["input"]["dnn_columns"]["categorical_columns_with_vocab"]:
         col = tf.feature_column.categorical_column_with_vocabulary_list(
@@ -114,7 +114,7 @@ def create_estimator(run_config, model_config):
         target_vocab = model_config["input"]["target_vocab"]
         num_classes = len(target_vocab)
 
-    return tf.estimator.DNNClassifier(
+    return tf.estimator.DNNLinearCombinedClassifier(
         linear_feature_columns=linear_feature_columns,
         dnn_feature_columns=dnn_feature_columns,
         n_classes=num_classes,

examples/iris/resources/models.yaml

@@ -5,7 +5,7 @@ def create_estimator(run_config, model_config):
     dnn_feature_columns = []
 
     for col_name in model_config["input"]["dnn_columns"]["numeric_columns"]:
-        feature_columns.append(tf.feature_column.numeric_column(col_name))
+        dnn_feature_columns.append(tf.feature_column.numeric_column(col_name))
 
     for col_info in model_config["input"]["dnn_columns"]["categorical_columns_with_vocab"]:
         col = tf.feature_column.categorical_column_with_vocabulary_list(
@@ -101,7 +101,7 @@ def create_estimator(run_config, model_config):
             )
         )
 
-    return tf.estimator.DNNClassifier(
+    return tf.estimator.DNNLinearCombinedRegressor(
         linear_feature_columns=linear_feature_columns,
         dnn_feature_columns=dnn_feature_columns,
         dnn_hidden_units=model_config["hparams"]["dnn_hidden_units"],

examples/iris/resources/raw_columns.yaml

@@ -57,7 +57,7 @@ def create_estimator(run_config, model_config):
         target_vocab = model_config["input"]["target_vocab"]
         num_classes = len(target_vocab)
 
-    return tf.estimator.DNNClassifier(
+    return tf.estimator.LinearClassifier(
         feature_columns=feature_columns,
         n_classes=num_classes,
         label_vocabulary=target_vocab,

examples/iris/resources/transformed_columns.yaml

@@ -44,7 +44,7 @@ def create_estimator(run_config, model_config):
             )
         )
 
-    return tf.estimator.DNNRegressor(
+    return tf.estimator.LinearRegressor(
         feature_columns=feature_columns,
         weight_column=model_config["input"].get("weight_column", None),
         config=run_config,