Add AvgPool3D layer #99
Conversation
| */ | ||
| public class AvgPool3D( | ||
| public val poolSize: IntArray = intArrayOf(2, 2, 2), | ||
| public val strides: IntArray? = null, |
There was a problem hiding this comment.
set the default value for the strides and convert both, strides and poolSize to the 5d array of longs
There was a problem hiding this comment.
Conversion was done, and I have mentioned the benefit of using null as the default value for strides.
api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/core/layer/pooling/AvgPool3D.kt
Outdated
Show resolved
Hide resolved
api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/core/layer/pooling/AvgPool3D.kt
Outdated
Show resolved
Hide resolved
api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/core/layer/pooling/AvgPool3D.kt
Outdated
Show resolved
Hide resolved
api/src/test/kotlin/org/jetbrains/kotlinx/dl/api/core/layer/AvgPool3DTest.kt
Outdated
Show resolved
Hide resolved
|
@zaleslaw I just merged it with master branch and updated it with recent/requested changes. |
api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/core/layer/pooling/AvgPool3D.kt
Outdated
Show resolved
Hide resolved
api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/core/layer/pooling/AvgPool3D.kt
Outdated
Show resolved
Hide resolved
| arrayOf( | ||
| arrayOf( | ||
| arrayOf( | ||
| floatArrayOf(18.0f/8, 4.5f/8, 16.0f/8), |
There was a problem hiding this comment.
did you take these values from Keras tests?
There was a problem hiding this comment.
No, I created the test data myself :)
There was a problem hiding this comment.
Not bad:)
Sometimes I just copied test data from Keras; sometimes, if test data seems not appropriate, I use Keras like a test machine writing some simple test in Python to put some test data in inputs and get outputs. A few times, I ignored this method and just used tested primitive itself to get data, but it has a bug in implementation and, as a result, a wrong test:(( But I agree that test inputs/outputs for pooling could be calculated manually (just shared my experience)
There was a problem hiding this comment.
Yeah, actually I have created test data for all test cases I have written so far; but certainly it's not a sustainable approach and using another tested tool, like Keras or maybe even numpy, for generating test data seems to be a better and more reliable approach.
|
@zaleslaw I changed the strides type and default value per your request. |
zaleslaw
added
LGTM
Resolves #81. This PR adds support for AvgPool3D layer.
NOTE: as commented in the unit tests, currently it gives an error when
dataFormat = CHANNELS_FIRST.