Update docs

Author: deliahu

docs/applications/implementations/estimators.md

@@ -26,6 +26,8 @@ def create_estimator(run_config, model_config):
     pass
 ```
 
+See the [tf.estimator.RunConfig](https://www.tensorflow.org/api_docs/python/tf/estimator/RunConfig) and [tf.estimator.Estimator](https://www.tensorflow.org/api_docs/python/tf/estimator/Estimator) documentation for more information.
+
 ## Example
 
 ```python

docs/applications/resources/data-types.md

@@ -35,11 +35,11 @@ Output types are used to define the output types of aggregators and constants. T
 
 In addition, an output type may be a list of scalar types. This is denoted by a length-one list of any of the supported types. For example, `[INT]` represents a list of integers.
 
-An output type may also be a map containing these types. There are two types of maps: generic maps and fixed maps. **Generic maps** represent maps which may have any number of items. The types of the keys and values must match the declared types. For example, `{STRING: INT}` fits `{"San Francisco": -7, "Toronto": -4}`. **Fixed maps** represent maps which must define values for each of the pre-defined keys. For example: `{value1: INT, value2: FLOAT}` fits `{value1: 17, value2: 8.8}`.
+An output type may also be a map containing these types. There are two types of maps: generic maps and fixed maps. **Generic maps** represent maps which may have any number of items. The types of the keys and values must match the declared types. For example, `{STRING: INT}` supports `{"San Francisco": -7, "Toronto": -4}`. **Fixed maps** represent maps which must define values for each of the pre-defined keys. For example: `{mean: INT, stddev: FLOAT}` supports `{mean: 17, stddev: 8.8}`.
 
 The values in lists, generic maps, and fixed maps may be arbitrarily nested.
 
-These are all valid output types:
+Here are some valid output types:
 
 * `INT`
 * `[STRING]`
@@ -87,18 +87,18 @@ Input types are used to define the inputs to aggregators, transformers, and esti
 * `STRING`
 * `BOOL`
 
-Like with output types, input types may occur within lists, generic maps, and fixed maps.
+Like with output types, input types may occur within arbitrarily nested lists, generic maps, and fixed maps.
 
-Ambiguous input types are also supported, and are represented by joining types with `|`. For example, `INT_COLUMN|FLOAT_COLUMN` indicates that either a column of type `INT_COLUMN` or a column of type `FLOAT_COLUMN` may be used an the input. Any two or more types may be combined in this way (e.g. `INT|FLOAT|STRING` is supported). All permutations of ambiguous types are valid (e.g. `INT|FLOAT` and `FLOAT|INT` are equivalent). Column types and scalar types may not be combined (e.g. `INT|FLOAT_COLUMN` is not valid).
+Ambiguous input types are also supported, and are represented by joining types with `|`. For example, `INT_COLUMN|FLOAT_COLUMN` indicates that either a column of type `INT_COLUMN` or a column of type `FLOAT_COLUMN` may be used as the input. Any two or more types may be combined in this way (e.g. `INT|FLOAT|STRING` is supported). All permutations of ambiguous types are valid (e.g. `INT|FLOAT` and `FLOAT|INT` are equivalent). Column types and scalar types may not be combined (e.g. `INT|FLOAT_COLUMN` is not valid).
 
 ### Input Type Validations
 
 By default, all declared inputs are required. For example, if the input type is `{value1: INT, value2: FLOAT}`, both `value1` and `value2` must be provided (and cannot be `Null`). With Cortex, it is possible to declare inputs as optional, set default values, allow values to be `Null`, and specify minimum and maximum map/list lengths.
 
 To specify validation options, the "long form" input schema is used. In the long form, the input type is always a map, with the `_type` key specifying the type, and other keys (which all start with `_`) specifying the options. The available options are:
 
-* `_optional`: If set to `True`, allows the value to be missing from the input. Only applicable to values in maps.
-* `_default`: Specifies a default value to use if the value is missing from the input. Only applicable to values in maps. Setting `_defaut` implies `_optional: True`.
+* `_optional`: If set to `True`, allows the value to be missing from the input. This only applies to values in maps.
+* `_default`: Specifies a default value to use if the value is missing from the input. This only applies to values in maps. Setting `_defaut` implies `_optional: True`.
 * `_allow_null`: If set to `True`, allows the value to be explicitly set to `Null`.
 * `_min_count`: Specifies the minimum number of elements that must be in the list or map.
 * `_max_count`: Specifies the maximum number of elements that must be in the list or map.
@@ -135,7 +135,7 @@ input:
     _max_count: 100
 ```
 
-Input value (assuming `column1` is an `INT_COLUMN`, `constant1` is a `[STRING]`, and `aggregate1` is an `INT`):
+Input value (assuming `column1` is an `INT_COLUMN`, `constant1` is a `[STRING]`, and `aggregate1` is an `STRING`):
 
 ```yaml
 input:

docs/applications/resources/models.md

@@ -58,8 +58,6 @@ Train TensorFlow models at scale.
 
 See [Data Types](data-types.md) for details about input and output values. Note: the `target_column`, `input`, `training_input`, and `hparams` of the the aggregate must match the input types of the estimator (if specified).
 
-See the [tf.estimator.RunConfig](https://www.tensorflow.org/api_docs/python/tf/estimator/RunConfig) and [tf.estimator.EvalSpec](https://www.tensorflow.org/api_docs/python/tf/estimator/EvalSpec) documentation for more information.
-
 ## Example
 
 ```yaml

docs/applications/resources/overview.md

@@ -9,6 +9,7 @@ Cortex applications consist of declarative resource configuration written in YAM
 * [aggregate](aggregates.md)
 * [transformer](transformers.md)
 * [transformed_column](transformed-columns.md)
+* [estimator](estimators.md)
 * [model](models.md)
 * [api](apis.md)
 * [constant](constants.md)

docs/summary.md

@@ -20,6 +20,7 @@
   * [Aggregates](applications/resources/aggregates.md)
   * [Transformers](applications/resources/transformers.md)
   * [Transformed Columns](applications/resources/transformed-columns.md)
+  * [Estimators](applications/resources/estimators.md)
   * [Models](applications/resources/models.md)
   * [APIs](applications/resources/apis.md)
   * [Constants](applications/resources/constants.md)
@@ -30,7 +31,7 @@
 
   * [Aggregators](applications/implementations/aggregators.md)
   * [Transformers](applications/implementations/transformers.md)
-  * [Models](applications/implementations/models.md)
+  * [Estimators](applications/implementations/estimators.md)
 
 ### Advanced
 

docs/tutorial.md

@@ -22,10 +22,10 @@ Our goal is to build a web API that returns the type of iris given its measureme
 
 ```bash
 mkdir iris && cd iris
-touch app.yaml dnn.py irises.json
+touch app.yaml irises.json
 ```
 
-Cortex requires an `app.yaml` file which defines a single `app` resource. Other resources may be defined in arbitrarily named YAML files in the the directory which contains `app.yaml` or any subdirectories. For this example, we will define all of our resources in `app.yaml`.
+Cortex requires an `app.yaml` file which defines an `app` resource. Other resources may be defined in arbitrarily named YAML files in the the directory which contains `app.yaml` or any subdirectories. For this example, we will define all of our resources in `app.yaml`.
 
 Add to `app.yaml`:
 
@@ -49,11 +49,11 @@ Add to `app.yaml`:
     schema: [@sepal_length, @sepal_width, @petal_length, @petal_width, @class]
 ```
 
-Cortex will be able to read from any S3 bucket that your AWS credentials grant access to.
+Cortex is able to read from any S3 bucket that your AWS credentials grant access to.
 
 #### Define the model
 
-This configuration will generate a training dataset with the specified columns and train our classifier using the generated dataset. Here we're using TensorFlow's [DNNClassifier](https://www.tensorflow.org/api_docs/python/tf/estimator/DNNClassifier) but Cortex supports any TensorFlow code that adheres to the [tf.estimator API](https://www.tensorflow.org/guide/estimators).
+This configuration will generate a training dataset with the specified columns and train our classifier using the generated dataset. Here we're using a built-in estimator (which uses TensorFlow's [DNNClassifier](https://www.tensorflow.org/api_docs/python/tf/estimator/DNNClassifier)) but Cortex supports any TensorFlow code that adheres to the [tf.estimator API](https://www.tensorflow.org/guide/estimators).
 
 Add to `app.yaml`:
 
@@ -87,7 +87,7 @@ Add to `app.yaml`:
   name: iris-type
   model: @dnn
   compute:
-    replicas: 1
+    replicas: 2
 ```
 
 ## Deploy the application

examples/external-model/app.yaml

@@ -1,5 +1,5 @@
 - kind: app
-  name: iris
+  name: external-model
 
 - kind: api
   name: iris