import copy
import numpy as np
import akida
from . import common
def _get_weights_params_identity(layer):
"""
Creates an 'identity' convolutional layer parameters and its weights.
"""
out_dims = layer.output_dims
nb_chan = out_dims[2]
dw_weights = np.zeros((3, 3, nb_chan, 1), dtype=np.int8)
pw_weights = np.zeros((1, 1, nb_chan, nb_chan), dtype=np.int8)
for i in range(nb_chan):
dw_weights[1, 1, i, 0] = 1
pw_weights[0, 0, i, i] = 1
# create a layer to have default parameters
identity_layer = akida.SeparableConvolutional(
name=f"{layer.name}_pooling",
kernel_size=(3, 3),
filters=nb_chan,
threshold=0,
act_bits=layer.parameters.act_bits,
act_step=2**layer.parameters.act_bits / 16)
return copy.copy(identity_layer.parameters), dw_weights, pw_weights
def _copy_layer_variables(layer, copied_layer):
for var in copied_layer.get_variable_names():
layer.set_variable(var, copied_layer.get_variable(var))
def _copy_layer(model, layer):
new_layer = akida.Layer(layer.parameters, layer.name)
model.add(new_layer)
if layer.learning:
# Recompile model with layer parameters
learn_params = {
attr: getattr(layer.learning, attr)
for attr in dir(layer.learning)
if '__' not in attr and 'learning_type' not in attr
}
model.compile(**learn_params)
_copy_layer_variables(new_layer, layer)
def _add_identity_cnp_after_max_pooling(model, layer):
"""
Adds the layer and an identity CNP to the model
"""
ident_params, ident_dw_weights, ident_pw_weights = _get_weights_params_identity(
layer)
identity_layer = akida.Layer(ident_params, f"{layer.name}_identity")
model.add(identity_layer)
identity_layer.set_variable("weights", ident_dw_weights)
identity_layer.set_variable("weights_pw", ident_pw_weights)
def _cnp_max_pooling(layer):
return layer.parameters.layer_type in [
akida.LayerType.Convolutional, akida.LayerType.SeparableConvolutional
] and layer.parameters.pool_type == akida.PoolType.Max
def _cnp_sep_avg_pooling(layer):
return (layer.parameters.layer_type
== akida.LayerType.SeparableConvolutional and
layer.parameters.pool_type == akida.PoolType.Average)
def _cnp_pooling_needs_identity_cnp(model, layer_index):
"""
Returns True if the layer is CNP with max pooling not followed by another
CNP, and we can add an identity CNP layer after it without altering result
"""
result = False
layer = model.get_layer(layer_index)
if _cnp_max_pooling(layer):
# if it is not the last layer, check the layer is not followed by
# another cnp
if layer_index != model.get_layer_count() - 1:
next_layer = model.get_layer(layer_index + 1)
if next_layer.parameters.layer_type not in [
akida.LayerType.Convolutional,
akida.LayerType.SeparableConvolutional
]:
result = True
# if it is the last layer, we can add an identity layer only if it has
# activations enabled
elif layer.parameters.activation:
result = True
return result
def _cnp_max_pooling_split(model, layer):
"""
Splits a CNP with max pooling into 2 CNPs:
- one performing the convolution
- the other one performing the pooling
"""
# 1st layer is the conv without pooling
conv_params = copy.copy(layer.parameters)
conv_params.pool_type = akida.PoolType.NoPooling
conv_params.pool_size = (-1, -1)
conv_params.pool_stride = (-1, -1)
layer_conv = akida.Layer(conv_params, f"{layer.name}_conv")
model.add(layer_conv)
_copy_layer_variables(layer_conv, layer)
# 2nd layer is an identity conv with pooling
pool_params, pool_dw_weights, pool_pw_weights = _get_weights_params_identity(
layer)
pool_params.pool_type = akida.PoolType.Max
pool_params.pool_size = layer.parameters.pool_size
pool_params.pool_stride = layer.parameters.pool_stride
pool_layer = akida.Layer(pool_params, f"{layer.name}_pooling")
model.add(pool_layer)
pool_layer.set_variable("weights", pool_dw_weights)
pool_layer.set_variable("weights_pw", pool_pw_weights)
def _cnp_sep_avg_pooling_add_dummy_neurons(model, layer):
"""
Performs compatibility check on SeparableConvolutional with
global average pooling:
- Adds dummy neurons if filters number is not a multiple of 8
- Adds 8 more dummy neurons to ensure all filters will be processed
"""
# Check if layer needs more neurons to be a multiple of 8
neurons_to_add = 8
if layer.parameters.filters % 8 != 0:
neurons_to_add += 8 - (layer.parameters.filters % 8)
# Add new layer w/ layer parameters
new_layer = akida.Layer(layer.parameters, layer.name)
new_layer.parameters.filters += neurons_to_add
model.add(new_layer)
# Copy variables and add dummy neurons
for var in layer.get_variable_names():
value = layer.get_variable(var)
if var == 'threshold':
new_value = np.concatenate(
(value, np.zeros(neurons_to_add, dtype=np.int32)))
elif var == 'act_step':
new_value = np.concatenate(
(value, np.ones(neurons_to_add, dtype=np.float32)))
elif var == 'weights_pw':
shape = value.shape
new_shape = (shape[0], shape[1], shape[2], neurons_to_add)
new_value = np.concatenate(
(value, np.zeros(new_shape, dtype=np.int8)), axis=3)
else:
# Copy dw weights
new_value = value
new_layer.set_variable(var, new_value)
return neurons_to_add
def _cnp_sep_avg_pooling_clone_extend_layer(model, layer, channels_to_add):
"""
Clone layer into the target model and append dummy channels
"""
# Create new layer
new_layer = akida.Layer(layer.parameters, layer.name)
model.add(new_layer)
for var in layer.get_variable_names():
value = layer.get_variable(var)
if var == 'weights':
shape = value.shape
new_shape = (shape[0], shape[1], channels_to_add, shape[3])
if layer.parameters.layer_type == akida.LayerType.FullyConnected:
compat_array = np.zeros(new_shape, dtype=np.int8)
else:
compat_array = np.ones(new_shape, dtype=np.int8)
new_value = np.concatenate((value, compat_array), axis=2)
elif var == 'weights_pw':
shape = value.shape
new_shape = (shape[0], shape[1], channels_to_add, shape[3])
new_value = np.concatenate(
(value, np.zeros(new_shape, dtype=np.int8)), axis=2)
else:
new_value = value
new_layer.set_variable(var, new_value)
[docs]def create_from_model(model, hw_version=None):
"""Tries to create a HW compatible model from an incompatible one
Tries to create a HW compatible model from an incompatible one, using SW
workarounds for known limitations. It returns a converted model that is not
guaranteed to be HW compatible, depending if workaround have been found.
Args:
model (:obj:`Model`): a Model object to convert
hw_version (:obj:`HwVersion`, optional): version of the Hardware
Returns:
:obj:`Model`: a new Model with no guarantee that it is HW compatible.
"""
added_neurons = 0
new_model = akida.Model()
nb_layers = model.get_layer_count()
for i in range(nb_layers):
layer = model.get_layer(i)
if _cnp_max_pooling(layer):
# On hardware, any CNP with max pooling must be followed by a CNP
# (to perform vertical pooling). If not, an identity CNP layer is
# then added.
# Moreover, on nsoc-v1, CNP with max pooling and negative thresholds
# is not supported. To avoid this situation, any CNP with max
# pooling (whatever the thresholds) is split into 2 CNPs:
# - one performing the convolution
# - the other one performing the pooling
if (hw_version == akida.NSoC_v1 and
not common.cnp_is_identity(layer)):
_cnp_max_pooling_split(new_model, layer)
else:
_copy_layer(new_model, layer)
# If CNP has max pooling and is not followed by another CNP, we can
# add an identity CNP layer
if _cnp_pooling_needs_identity_cnp(model, i):
_add_identity_cnp_after_max_pooling(new_model, layer)
continue
# If CNP has an average pooling on a SeparableConvolutional, it has to
# be updated to add dummy neurons, except if this is the last layer
if hw_version == akida.NSoC_v1 and _cnp_sep_avg_pooling(
layer) and i != nb_layers - 1:
added_neurons = _cnp_sep_avg_pooling_add_dummy_neurons(
new_model, layer)
continue
if added_neurons > 0:
previous_layer = model.get_layer(i - 1)
# Check if previous layer was a SeparableConvolutional with
# global average pooling. If yes, current layer needs to be
# updated to be aligned with the previous one.
if _cnp_sep_avg_pooling(previous_layer):
# Following layer has to be a SeparableConvolutional or a
# FullyConnected. If not an error message will be raised.
if layer.parameters.layer_type != akida.LayerType.Convolutional:
_cnp_sep_avg_pooling_clone_extend_layer(
new_model, layer, added_neurons)
continue
raise TypeError(
'SeperableConvolutional with average pooling cannot be '
'followed by a Convolutional layer')
# if no particular case is found, copy the layer into the new model
_copy_layer(new_model, layer)
return new_model