Revise section on Host and Device Arrays

Author: csadorf

wiki/python/ESTIMATOR_GUIDE.md

@@ -162,24 +162,34 @@ Internally, all arrays should be converted to `CumlArray` as much as possible si
 
 ### Host and Device Arrays
 
-Beginning with version 23.02, cuML provides support for executing at least
-some algorithms either on CPU or on GPU. Therefore, `CumlArray` objects can
-now be backed by either host or device memory.
-
-To ensure that arrays used by algorithms are backed by the correct memory
-type, two new global settings were introduced: `device_type` (`'cpu'` or
-`'gpu'`) and `memory_type` (`'host'` or `'device'`). The former indicates
-what sort of computational device will be used to execute an algorithm while
-the latter indicates where arrays should be stored if not otherwise
-specified. If the `device_type` is updated to a value incompatible with the
-current `memory_type`, the `memory_type` will be changed to something
-compatible with `device_type`, but the reverse is not true. This allows for
-e.g. allocating an array where results will ultimately be stored even if the
-actual computation will take place on a different device.
-
-New array output types were also introduced to take advantage of these
-settings by deferring where appropriate to the globally-set memory type. Read
-on for more details on how to take advantage of these types.
+cuML provides flexible memory management through the `CumlArray` class, which can store data in either device-accessible or host-accessible memory. This flexibility allows for efficient data handling and computation.
+
+To control where arrays are stored, the `memory_type` setting determines where arrays are stored by default. This setting can be configured through:
+
+1. Global settings:
+   ```python
+   import cuml
+   cuml.global_settings.memory_type = 'device'  # For device-accessible memory
+   ```
+
+2. Context managers:
+   ```python
+   with cuml.using_memory_type('device'):
+       # Arrays created here will use device-accessible memory
+       pass
+   ```
+
+3. Individual estimator settings:
+   ```python
+   estimator = MyEstimator(output_mem_type='device')
+   ```
+
+The memory type settings allow for efficient data management:
+- Use device-accessible memory for GPU-accelerated computations
+- Use host-accessible memory for data that needs to be accessed by the CPU
+- Let cuML handle memory transfers automatically when needed
+
+New array output types were introduced to take advantage of these settings by deferring to the globally-set memory type. Read on for more details on how to take advantage of these types.
 
 ### Specifying the Array Output Type