An attempt at a base class for .py encoders #831
Conversation
| import numpy as np | ||
| from htm.bindings.sdr import SDR | ||
|
|
||
| class BaseEncoder(ABC): |
There was a problem hiding this comment.
we've touched this..it's probably not worth creating a pybind interface for BaseEncoder.hpp? This file will be more flexible
|
|
||
| @abstractmethod | ||
| def reset(self): | ||
| raise NotImplementedError() |
There was a problem hiding this comment.
i mistified you about the raise, with @abstractmethod you should replace the raise with just pass
There was a problem hiding this comment.
...and also remove the @AbstractMethod, most encoders dont use reset, so they'll use this implementation
def reset(self):
pass
| @@ -20,8 +20,9 @@ | |||
|
|
|||
| from htm.bindings.sdr import SDR | |||
| from htm.bindings.math import Random | |||
| from htm.bindings.encoder import BaseEncoder | |||
There was a problem hiding this comment.
the above is not in htm.bindings
| class BaseEncoder(ABC): | ||
|
|
||
| @abstractmethod | ||
| def __init__(dimensions): |
There was a problem hiding this comment.
might think off of some common, useful params here: debug=0, inputType=None,...
| raise NotImplementedError() | ||
|
|
||
| @abstractmethod | ||
| def encode(self, input, output): |
There was a problem hiding this comment.
this one will remain abstract -> we want all subclasses to reimplement it.
But might add some common checks to save code elsewhere:
def encode(self, inp, output=None):
if output is None:
output = SDR(self.dimensions)
else:
assert( isinstance(output, SDR) )
assert( all(x == y for x, y in zip( output.dimensions, self.dimensions )))
if inp is None or (isinstance(inp, float) and math.isnan(inp)):
output.zero()
return output
| @abstractmethod | ||
| def __init__(dimensions): | ||
| self.dimensions = dimensions | ||
| self.size = SDR(dimensions).size |
| @@ -60,14 +61,14 @@ def __init__(self, | |||
| assert(self.sparsity <= 1) | |||
|
|
|||
| # Assign each module a range of cells in the output SDR. | |||
| partitions = np.linspace(0, self.size, num=len(self.periods) + 1) | |||
| partitions = np.linspace(0, super().size, num=len(self.periods) + 1) | |||
There was a problem hiding this comment.
subclass has all the members,methods of its superclass, so this could/should remain self.size, or even size
| @@ -50,8 +51,8 @@ def __init__(self, | |||
| encoder uses. This encoder produces deterministic output. The seed | |||
| zero is special, seed zero is replaced with a truly random seed. | |||
| """ | |||
| self.size = size | |||
| self.dimensions = (size,) | |||
| super().__init__((size,)) | |||
There was a problem hiding this comment.
👍 this is a correct use of explicitely calling the super's methods. (note: unlike in c++, you dont need to call the super.init as the first statement
| @@ -95,10 +96,10 @@ def encode(self, location, grid_cells=None): | |||
| location = list(location) | |||
There was a problem hiding this comment.
encode should call super().encode(location)
|
@breznak would you please finish this for me. Thanks. |
This is not complete...
..But I have to admit I don't know what I am doing. I am not proficient at Python to do the right thing here. @breznak can you help me?
What I am trying to do is develop a base class that can be used by any encoder developed in Python to ensure it implements the encode( ) function and sets the dimensions. I was trying to insert this into the existing grid_cell_encoder.py to make sure it works.