Comments (3)
The custom layer is the right idea.
ops.reshape
works with ops.shape
, but only with actual tensors -- not just symbolic tensors, that is to say Input
objects. That means that a tensor's shape can be always be known inside the call method of a layer, which is why querying ops.shape
and calling reshape
inside call()
works fine.
The reason it worked with tf.keras
is that in tf.keras
, an Input
is backed by a TF tensor. This is not the case with Keras Core, Input
is just a standalone Python object and its shape may contain None
entries.
from keras-core.
Just came up with a workaround after posing:
import keras_core
import tensorflow as tf
from keras_core import layers
"""
Workaround
"""
class CustomReshape(layers.Layer):
def __init__(self, name=None):
super().__init__(name=name)
def compute_output_shape(self, input_shape):
return (*input_shape, 1)
def call(self, x):
b, h, w, c = tf.shape(x)
return tf.reshape(x, (b, h, w, c, 1))
inputs = keras_core.layers.Input(shape=(None, None, 3))
x = CustomReshape()(inputs)
model = keras_core.models.Model(inputs=inputs, outputs=x)
x = tf.random.uniform(shape=(1, 28, 28, 3))
y = model(x)
print(y.shape)
from keras-core.
Thanks for the detailed clarification.
I provide the following backend-agnostic workaround in case anyone encounters this issue
import keras_core
from keras_core import layers
"""
Workaround: ops.reshape works with ops.shape with actual tensors
"""
class CustomReshape(layers.Layer):
def __init__(self, name=None):
super().__init__(name=name)
def compute_output_shape(self, input_shape):
return (*input_shape, 1)
def call(self, x):
b, h, w, c = keras_core.ops.shape(x)
return keras_core.ops.reshape(x, (b, h, w, c, 1))
inputs = keras_core.layers.Input(shape=(None, None, 3))
x = CustomReshape()(inputs)
model = keras_core.models.Model(inputs=inputs, outputs=x)
x = keras_core.ops.random.uniform(shape=(1, 28, 28, 3))
y = model(x)
print(y.shape)
from keras-core.
Related Issues (20)
- tflite cannot fuse `BatchNormalization` in Keras Core as effectively as in the original Keras
- Improve Documentation on KERAS_BACKEND HOT 5
- GSOC Program Organization ( question ) HOT 1
- Constant Initializer supports only Scalar values HOT 1
- Add support for RaggedTensors to JAX and Pytorch Backend HOT 1
- Saving broken between tf.keras and Keras Core HOT 9
- Ensure workflow reliability by hash-pinning GitHub Actions HOT 1
- Run actions.yml with read-only permissions
- Segmentation metrics HOT 3
- Use of PyTorch loss functions inside Keras
- Is possible serialize models which use torch functions? HOT 5
- Expose `Operation` without `src` import hacks HOT 1
- v0.1.6 bug: AttributeError: 'GPT2CausalLM' object has no attribute 'compiled' HOT 2
- Adam with amsgrad=True + JAX backend is broken HOT 1
- Inconsistent type handling between backends HOT 1
- Matmul - tensorflow does not broadcast/expand dimensions correctly HOT 3
- Casting dtype in losses.Loss base class HOT 2
- TypeError: copy() got an unexpected keyword argument 'overwrite' HOT 2
- Error when running the model.fit on the wrapped model "Only input tensors may be passed as positional arguments."
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from keras-core.