elda27 / prepnet Goto Github PK

The implementation of some converters can't reuse.
The example shows the following.

input1 = np.array([
    'Tom',
    'Becky',
    'Tom',
    'Becky',
    'Ann',
])
converter = prepnet.OrdinalConverter()
result = converter.encode(input1)
print(result) # Display text should be "[0, 1, 0, 1, 2]"

input2 = np.array([
    'Tom',
    'Tom',
    'Ann',
])
result = converter.encode(inputs)
print(result) # Display text should be "[0, 0, 2]" but actual "[0, 0, 1]"

Need enhance the examples

• Example1: Directly use converters.
• Example2: Functional based converters
• Example3: Column wise based converters (using FrameConverter)

Now the master branch is very complicated due to FrameConverterContext and Asynchronous Generator (encode_async and decode_async).
Implementations should be more simple for extensions.

These are rethinking implementation.

• Executor will split ColumnExecutor and FrameExecutor.
• Remove encode_async and decode_async
• Remove executor dependency such as FrameConverterContext.
• FunctionalContext should use above executors.
• Some tests change for above changes.

efc4810

Considering readme text.