DL: Update error msg for missing input_shape
JIRA: MADLIB-1438
Previously, `keras` would error out if the first layer didn't have an
input shape but with `tf.keras` there is way to create a layer without
an input shape.(https://keras.io/examples/vision/mnist_convnet/). In
this case, the model JSON does not have it either. We need the input
shape for validation and also we cannot set model weights without the
input shape being added to the first layer.
Co-authored-by: Nikhil Kak <nkak@vmware.com>
diff --git a/src/ports/postgres/modules/deep_learning/model_arch_info.py_in b/src/ports/postgres/modules/deep_learning/model_arch_info.py_in
index 3a225cc..8f7418b 100644
--- a/src/ports/postgres/modules/deep_learning/model_arch_info.py_in
+++ b/src/ports/postgres/modules/deep_learning/model_arch_info.py_in
@@ -39,7 +39,8 @@
arch_layers = _get_layers(model_arch)
if 'batch_input_shape' in arch_layers[0]['config']:
return arch_layers[0]['config']['batch_input_shape'][1:]
- plpy.error('Unable to get input shape from model architecture.')
+ plpy.error('Unable to get input shape from model architecture.'\
+ 'Make sure that the first layer defines an input_shape.')
def get_num_classes(model_arch):
"""
