conv.py 35.1 KB
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912
#!/usr/bin/env python
# -*- coding: utf-8 -*-

# Contributors : Titouan Parcollet
# Initial Authors: Chiheb Trabelsi

from keras import backend as K
from keras import activations, initializers, regularizers, constraints
from keras.layers import Lambda, Layer, InputSpec, Convolution1D, Convolution2D, add, multiply, Activation, Input, concatenate
from keras.layers.convolutional import _Conv
from keras.layers.merge import _Merge
from keras.layers.recurrent import Recurrent
from keras.utils import conv_utils
from keras.models import Model
import numpy as np
from theano.sandbox.rng_mrg import MRG_RandomStreams as RandomStreams
from .fft import fft, ifft, fft2, ifft2
from .init import *
import sys


#####################################################################
#Quaternion Implementations					    #
#####################################################################

class QuaternionConv(Layer):
	"""Abstract nD quaternion convolution layer.
	This layer creates a quaternion convolution kernel that is convolved
	with the layer input to produce a tensor of outputs.
	If `use_bias` is True, a bias vector is created and added to the outputs.
	Finally, if `activation` is not `None`,
	it is applied to the outputs as well.
	# Arguments
		rank: An integer, the rank of the convolution,
			e.g. "2" for 2D convolution.
		filters: Integer, the dimensionality of the output space, i.e,
			the number of quaternion feature maps. It is also the effective number
			of feature maps for each of the real and imaginary parts.
			(i.e. the number of quaternion filters in the convolution)
			The total effective number of filters is 2 x filters.
		kernel_size: An integer or tuple/list of n integers, specifying the
			dimensions of the convolution window.
		strides: An integer or tuple/list of n integers,
			spfying the strides of the convolution.
			Specifying any stride value != 1 is incompatible with specifying
			any `dilation_rate` value != 1.
		padding: One of `"valid"` or `"same"` (case-insensitive).
		data_format: A string,
			one of `channels_last` (default) or `channels_first`.
			The ordering of the dimensions in the inputs.
			`channels_last` corresponds to inputs with shape
			`(batch, ..., channels)` while `channels_first` corresponds to
			inputs with shape `(batch, channels, ...)`.
			It defaults to the `image_data_format` value found in your
			Keras config file at `~/.keras/keras.json`.
			If you never set it, then it will be "channels_last".
		dilation_rate: An integer or tuple/list of n integers, specifying
			the dilation rate to use for dilated convolution.
			Currently, specifying any `dilation_rate` value != 1 is
			incompatible with specifying any `strides` value != 1.
		activation: Activation function to use
			(see keras.activations).
			If you don't specify anything, no activation is applied
			(ie. "linear" activation: `a(x) = x`).
		use_bias: Boolean, whether the layer uses a bias vector.
		normalize_weight: Boolean, whether the layer normalizes its quaternion
			weights before convolving the quaternion input.
			The quaternion normalization performed is similar to the one
			for the batchnorm. Each of the quaternion kernels are centred and multiplied by
			the inverse square root of covariance matrix.
			Then, a quaternion multiplication is perfromed as the normalized weights are
			multiplied by the quaternion scaling factor gamma.
		kernel_initializer: Initializer for the quaternion `kernel` weights matrix.
			By default it is 'quaternion'. The 'quaternion_independent' 
			and the usual initializers could also be used.
			(see keras.initializers and init.py).
		bias_initializer: Initializer for the bias vector
			(see keras.initializers).
		kernel_regularizer: Regularizer function applied to
			the `kernel` weights matrix
			(see keras.regularizers).
		bias_regularizer: Regularizer function applied to the bias vector
			(see keras.regularizers).
		activity_regularizer: Regularizer function applied to
			the output of the layer (its "activation").
			(see keras.regularizers).
		kernel_constraint: Constraint function applied to the kernel matrix
			(see keras.constraints).
		bias_constraint: Constraint function applied to the bias vector
			(see keras.constraints).
		spectral_parametrization: Whether or not to use a spectral
			parametrization of the parameters.
	"""

	def __init__(self, rank,
				 filters,
				 kernel_size,
				 strides=1,
				 padding='valid',
				 data_format=None,
				 dilation_rate=1,
				 activation=None,
				 use_bias=True,
				 normalize_weight=False,
				 kernel_initializer='quaternion',
				 bias_initializer='zeros',
				 gamma_diag_initializer=sqrt_init,
				 gamma_off_initializer='zeros',
				 kernel_regularizer=None,
				 bias_regularizer=None,
				 gamma_diag_regularizer=None,
				 gamma_off_regularizer=None,
				 activity_regularizer=None,
				 kernel_constraint=None,
				 bias_constraint=None,
				 gamma_diag_constraint=None,
				 gamma_off_constraint=None,
				 init_criterion='he',
				 seed=None,
				 spectral_parametrization=False,
				 epsilon=1e-7,
				 **kwargs):
		super(QuaternionConv, self).__init__(**kwargs)
		self.rank = rank
		self.filters = filters
		self.kernel_size = conv_utils.normalize_tuple(kernel_size, rank, 'kernel_size')
		self.strides = conv_utils.normalize_tuple(strides, rank, 'strides')
		self.padding = conv_utils.normalize_padding(padding)
		self.data_format = 'channels_last' if rank == 1 else conv_utils.normalize_data_format(data_format)
		self.dilation_rate = conv_utils.normalize_tuple(dilation_rate, rank, 'dilation_rate')
		self.activation = activations.get(activation)
		self.use_bias = use_bias
		self.normalize_weight = normalize_weight
		self.init_criterion = init_criterion
		self.spectral_parametrization = spectral_parametrization
		self.epsilon = epsilon
		self.kernel_initializer = sanitizedInitGet(kernel_initializer)
		self.bias_initializer = sanitizedInitGet(bias_initializer)
		self.gamma_diag_initializer = sanitizedInitGet(gamma_diag_initializer)
		self.gamma_off_initializer = sanitizedInitGet(gamma_off_initializer)
		self.kernel_regularizer = regularizers.get(kernel_regularizer)
		self.bias_regularizer = regularizers.get(bias_regularizer)
		self.gamma_diag_regularizer = regularizers.get(gamma_diag_regularizer)
		self.gamma_off_regularizer = regularizers.get(gamma_off_regularizer)
		self.activity_regularizer = regularizers.get(activity_regularizer)
		self.kernel_constraint = constraints.get(kernel_constraint)
		self.bias_constraint = constraints.get(bias_constraint)
		self.gamma_diag_constraint = constraints.get(gamma_diag_constraint)
		self.gamma_off_constraint = constraints.get(gamma_off_constraint)
		if seed is None:
			self.seed = np.random.randint(1, 10e6)
		else:
			self.seed = seed
		self.input_spec = InputSpec(ndim=self.rank + 2)

	def build(self, input_shape):

		if self.data_format == 'channels_first':
			channel_axis = 1
		else:
			channel_axis = -1
		
		if input_shape[channel_axis] is None:
			raise ValueError('The channel dimension of the inputs '
							 'should be defined. Found `None`.')
		
		input_dim = input_shape[channel_axis] // 4
		self.kernel_shape = self.kernel_size + (input_dim , self.filters)
		# The kernel shape here is a complex kernel shape:
		#   nb of complex feature maps = input_dim;
		#   nb of output complex feature maps = self.filters;
		#   imaginary kernel size = real kernel size 
		#						 = self.kernel_size 
		#						 = complex kernel size
		if self.kernel_initializer in {'quaternion', 'quaternion_independent'}:
			kls = {'quaternion':			 QuaternionInit,
				   'quaternion_independent': QuaternionIndependentFilters}[self.kernel_initializer]
			kern_init = kls(
				kernel_size=self.kernel_size,
				input_dim=input_dim,
				weight_dim=self.rank,
				nb_filters=self.filters,
				criterion=self.init_criterion
			)
		else:
			kern_init = self.kernel_initializer
		
		self.kernel = self.add_weight(
			self.kernel_shape,
			initializer=kern_init,
			name='kernel',
			regularizer=self.kernel_regularizer,
			constraint=self.kernel_constraint
		)
		
		if self.normalize_weight:
			gamma_shape = (input_dim * self.filters,)
			self.gamma_rr = self.add_weight(
				shape=gamma_shape,
				name='gamma_rr',
				initializer=self.gamma_diag_initializer,
				regularizer=self.gamma_diag_regularizer,
				constraint=self.gamma_diag_constraint
			)

			self.gamma_ri = self.add_weight(
				shape=gamma_shape,
				name='gamma_ri',
				initializer=self.gamma_off_initializer,
				regularizer=self.gamma_off_regularizer,
				constraint=self.gamma_off_constraint
			)
			self.gamma_rj = self.add_weight(
				shape=gamma_shape,
				name='gamma_rj',
				initializer=self.gamma_off_initializer,
				regularizer=self.gamma_off_regularizer,
				constraint=self.gamma_off_constraint
			)
			self.gamma_rk = self.add_weight(
				shape=gamma_shape,
				name='gamma_rk',
				initializer=self.gamma_off_initializer,
				regularizer=self.gamma_off_regularizer,
				constraint=self.gamma_off_constraint
			)
			self.gamma_ii = self.add_weight(
				shape=gamma_shape,
				name='gamma_ii',
				initializer=self.gamma_diag_initializer,
				regularizer=self.gamma_diag_regularizer,
				constraint=self.gamma_diag_constraint
			)

			self.gamma_ij = self.add_weight(
				shape=gamma_shape,
				name='gamma_ij',
				initializer=self.gamma_off_initializer,
				regularizer=self.gamma_off_regularizer,
				constraint=self.gamma_off_constraint
			)
			self.gamma_ik = self.add_weight(
				shape=gamma_shape,
				name='gamma_ik',
				initializer=self.gamma_off_initializer,
				regularizer=self.gamma_off_regularizer,
				constraint=self.gamma_off_constraint
			)
			self.gamma_jj = self.add_weight(
				shape=gamma_shape,
				name='gamma_jj',
				initializer=self.gamma_diag_initializer,
				regularizer=self.gamma_diag_regularizer,
				constraint=self.gamma_diag_constraint
			)
			self.gamma_jk = self.add_weight(
				shape=gamma_shape,
				name='gamma_jk',
				initializer=self.gamma_diag_initializer,
				regularizer=self.gamma_diag_regularizer,
				constraint=self.gamma_diag_constraint
			)
			self.gamma_kk = self.add_weight(
				shape=gamma_shape,
				name='gamma_kk',
				initializer=self.gamma_off_initializer,
				regularizer=self.gamma_off_regularizer,
				constraint=self.gamma_off_constraint
			)
		else:
			self.gamma_rr = None
			self.gamma_ri = None
			self.gamma_rj = None
			self.gamma_rk = None
			self.gamma_ii = None
			self.gamma_ij = None
			self.gamma_ik = None
			self.gamma_jj = None
			self.gamma_jk = None
			self.gamma_kk = None
		
		#End of non understanded block

		if self.use_bias:
			bias_shape = (4 * self.filters,)
			self.bias = self.add_weight(
				bias_shape,
				initializer=self.bias_initializer,
				name='bias',
				regularizer=self.bias_regularizer,
				constraint=self.bias_constraint
			)

		else:
			self.bias = None

		# Set input spec.
		self.input_spec = InputSpec(ndim=self.rank + 2,
									axes={channel_axis: input_dim * 4})
		self.built = True

	def call(self, inputs):
		channel_axis = 1 if self.data_format == 'channels_first' else -1
		input_dim	= K.shape(inputs)[channel_axis] // 4
		index2 = self.filters*2
		index3 = self.filters*3
		if self.rank == 1:
			f_r   = self.kernel[:, :, :self.filters]
			f_i   = self.kernel[:, :, self.filters:index2]
			f_j   = self.kernel[:, :, index2:index3]
			f_k   = self.kernel[:, :, index3:]
		elif self.rank == 2:
			f_r   = self.kernel[:, :, :, :self.filters]
			f_i   = self.kernel[:, :, :, self.filters:index2]
			f_j   = self.kernel[:, :, :, index2:index3]
			f_k   = self.kernel[:, :, :, index3:]
		elif self.rank == 3:
			f_r   = self.kernel[:, :, :, :, :self.filters]
			f_i   = self.kernel[:, :, :, :, self.filters:index2]
			f_j   = self.kernel[:, :, :, :, index2:index3]
			f_k   = self.kernel[:, :, :, :, index3:]

		convArgs = {"strides":	   self.strides[0]	   if self.rank == 1 else self.strides,
					"padding":	   self.padding,
					"data_format":   self.data_format,
					"dilation_rate": self.dilation_rate[0] if self.rank == 1 else self.dilation_rate}
		convFunc = {1: K.conv1d,
					2: K.conv2d,
					3: K.conv3d}[self.rank]

		# processing if the weights are assumed to be represented in the spectral domain
		# Do we conserve this for quaternions ? Currently no

		if self.spectral_parametrization:
			print("Quaternion spectral weights parametrization not implemented yet, aborting.")
			sys.exit(1)
			if   self.rank == 1:
				f_r = K.permute_dimensions(f_r, (2,1,0))
				f_i = K.permute_dimensions(f_i, (2,1,0))
				f	  = K.concatenate([f_r, f_i], axis=0)
				fshape = K.shape(f)
				f	  = K.reshape(f, (fshape[0] * fshape[1], fshape[2]))
				f	  = ifft(f)
				f	  = K.reshape(f, fshape)
				f_r = f[:fshape[0]//2]
				f_i = f[fshape[0]//2:]
				f_r = K.permute_dimensions(f_r, (2,1,0))
				f_i = K.permute_dimensions(f_i, (2,1,0))
			elif self.rank == 2:
				f_r = K.permute_dimensions(f_r, (3,2,0,1))
				f_i = K.permute_dimensions(f_i, (3,2,0,1))
				f	  = K.concatenate([f_r, f_i], axis=0)
				fshape = K.shape(f)
				f	  = K.reshape(f, (fshape[0] * fshape[1], fshape[2], fshape[3]))
				f	  = ifft2(f)
				f	  = K.reshape(f, fshape)
				f_r = f[:fshape[0]//2]
				f_i = f[fshape[0]//2:]
				f_r = K.permute_dimensions(f_r, (2,3,1,0))
				f_i = K.permute_dimensions(f_i, (2,3,1,0))

		# In case of weight normalization, real and imaginary weights are normalized

		if self.normalize_weight:
			
			print("Quaternion weights normalization not implemented yet, aborting.")
			sys.exit(1)
			ker_shape = self.kernel_shape
			nb_kernels = ker_shape[-2] * ker_shape[-1]
			kernel_shape_4_norm = (np.prod(self.kernel_size), nb_kernels)
			reshaped_f_r = K.reshape(f_r, kernel_shape_4_norm)
			reshaped_f_i = K.reshape(f_i, kernel_shape_4_norm)
			reduction_axes = list(range(2))
			del reduction_axes[-1]
			mu_real = K.mean(reshaped_f_r, axis=reduction_axes)
			mu_imag = K.mean(reshaped_f_i, axis=reduction_axes)

			broadcast_mu_shape = [1] * 2
			broadcast_mu_shape[-1] = nb_kernels
			broadcast_mu_real = K.reshape(mu_real, broadcast_mu_shape)
			broadcast_mu_imag = K.reshape(mu_imag, broadcast_mu_shape)
			reshaped_f_r_centred = reshaped_f_r - broadcast_mu_real
			reshaped_f_i_centred = reshaped_f_i - broadcast_mu_imag
			Vrr = K.mean(reshaped_f_r_centred ** 2, axis=reduction_axes) + self.epsilon
			Vii = K.mean(reshaped_f_i_centred ** 2, axis=reduction_axes) + self.epsilon
			Vri = K.mean(reshaped_f_r_centred * reshaped_f_i_centred,
						 axis=reduction_axes) + self.epsilon
			
			normalized_weight = complex_normalization(
				K.concatenate([reshaped_f_r, reshaped_f_i], axis=-1),
				Vrr, Vii, Vri,
				beta = None,
				gamma_rr = self.gamma_rr,
				gamma_ri = self.gamma_ri,
				gamma_ii = self.gamma_ii,
				scale=True,
				center=False,
				axis=-1
			)

			normalized_real = normalized_weight[:, :nb_kernels]
			normalized_imag = normalized_weight[:, nb_kernels:]
			f_r = K.reshape(normalized_real, self.kernel_shape)
			f_i = K.reshape(normalized_imag, self.kernel_shape)
		
		#
		# Performing quaternion convolution
		#
		
		f_r._keras_shape = self.kernel_shape
		f_i._keras_shape = self.kernel_shape
		f_j._keras_shape = self.kernel_shape
		f_k._keras_shape = self.kernel_shape

		cat_kernels_4_r = K.concatenate([f_r, -f_i, -f_j, -f_k], axis=-2)
		cat_kernels_4_i = K.concatenate([f_i, f_r, -f_k, f_j], axis=-2)
		cat_kernels_4_j = K.concatenate([f_j, f_k, f_r, -f_i], axis=-2)
		cat_kernels_4_k = K.concatenate([f_k, -f_j, f_i, f_r], axis=-2)

		cat_kernels_4_quaternion = K.concatenate([cat_kernels_4_r, cat_kernels_4_i, cat_kernels_4_j, cat_kernels_4_k], axis=-1)
		cat_kernels_4_quaternion._keras_shape = self.kernel_size + (4 * input_dim, 4 * self.filters)

		output = convFunc(inputs, cat_kernels_4_quaternion, **convArgs)

		if self.use_bias:
			output = K.bias_add(
				output,
				self.bias,
				data_format=self.data_format
			)

		if self.activation is not None:
			output = self.activation(output)

		return output

	def compute_output_shape(self, input_shape):
		if self.data_format == 'channels_last':
			space = input_shape[1:-1]
			new_space = []
			for i in range(len(space)):
				new_dim = conv_utils.conv_output_length(
					space[i],
					self.kernel_size[i],
					padding=self.padding,
					stride=self.strides[i],
					dilation=self.dilation_rate[i]
				)
				new_space.append(new_dim)
			return (input_shape[0],) + tuple(new_space) + (4 * self.filters,)
		if self.data_format == 'channels_first':
			space = input_shape[2:]
			new_space = []
			for i in range(len(space)):
				new_dim = conv_utils.conv_output_length(
					space[i],
					self.kernel_size[i],
					padding=self.padding,
					stride=self.strides[i],
					dilation=self.dilation_rate[i])
				new_space.append(new_dim)
			return (input_shape[0],) + (4 * self.filters,) + tuple(new_space)

	def get_config(self):
		config = {
			'rank': self.rank,
			'filters': self.filters,
			'kernel_size': self.kernel_size,
			'strides': self.strides,
			'padding': self.padding,
			'data_format': self.data_format,
			'dilation_rate': self.dilation_rate,
			'activation': activations.serialize(self.activation),
			'use_bias': self.use_bias,
			'normalize_weight': self.normalize_weight,
			'kernel_initializer': sanitizedInitSer(self.kernel_initializer),
			'bias_initializer': sanitizedInitSer(self.bias_initializer),
			'gamma_diag_initializer': sanitizedInitSer(self.gamma_diag_initializer),
			'gamma_off_initializer': sanitizedInitSer(self.gamma_off_initializer),
			'kernel_regularizer': regularizers.serialize(self.kernel_regularizer),
			'bias_regularizer': regularizers.serialize(self.bias_regularizer),
			'gamma_diag_regularizer': regularizers.serialize(self.gamma_diag_regularizer),
			'gamma_off_regularizer': regularizers.serialize(self.gamma_off_regularizer),
			'activity_regularizer': regularizers.serialize(self.activity_regularizer),
			'kernel_constraint': constraints.serialize(self.kernel_constraint),
			'bias_constraint': constraints.serialize(self.bias_constraint),
			'gamma_diag_constraint': constraints.serialize(self.gamma_diag_constraint),
			'gamma_off_constraint': constraints.serialize(self.gamma_off_constraint),
			'init_criterion': self.init_criterion,
			'spectral_parametrization': self.spectral_parametrization,
		}
		base_config = super(QuaternionConv, self).get_config()
		return dict(list(base_config.items()) + list(config.items()))



class QuaternionConv1D(QuaternionConv):
	"""1D quaternion convolution layer.
	This layer creates a quaternion convolution kernel that is convolved
	with a quaternion input layer over a single quaternion spatial (or temporal) dimension
	to produce a quaternion output tensor.
	If `use_bias` is True, a bias vector is created and added to the quaternion output.
	Finally, if `activation` is not `None`,
	it is applied each of the real and imaginary parts of the output.
	When using this layer as the first layer in a model,
	provide an `input_shape` argument
	(tuple of integers or `None`, e.g.
	`(10, 128)` for sequences of 10 vectors of 128-dimensional vectors,
	or `(None, 128)` for variable-length sequences of 128-dimensional vectors.
	# Arguments
		filters: Integer, the dimensionality of the output space, i.e,
			the number of quaternion feature maps. It is also the effective number
			of feature maps for each of the real and imaginary parts.
			(i.e. the number of quaternion filters in the convolution)
			The total effective number of filters is 2 x filters.
		kernel_size: An integer or tuple/list of n integers, specifying the
			dimensions of the convolution window.
		strides: An integer or tuple/list of a single integer,
			specifying the stride length of the convolution.
			Specifying any stride value != 1 is incompatible with specifying
			any `dilation_rate` value != 1.
		padding: One of `"valid"`, `"causal"` or `"same"` (case-insensitive).
			`"causal"` results in causal (dilated) convolutions, e.g. output[t]
			does not depend on input[t+1:]. Useful when modeling temporal data
			where the model should not violate the temporal order.
			See [WaveNet: A Generative Model for Raw Audio, section 2.1](https://arxiv.org/abs/1609.03499).
		dilation_rate: an integer or tuple/list of a single integer, specifying
			the dilation rate to use for dilated convolution.
			Currently, specifying any `dilation_rate` value != 1 is
			incompatible with specifying any `strides` value != 1.
		activation: Activation function to use
			(see keras.activations).
			If you don't specify anything, no activation is applied
			(ie. "linear" activation: `a(x) = x`).
		use_bias: Boolean, whether the layer uses a bias vector.
		normalize_weight: Boolean, whether the layer normalizes its quaternion
			weights before convolving the quaternion input.
			The quaternion normalization performed is similar to the one
			for the batchnorm. Each of the quaternion kernels are centred and multiplied by
			the inverse square root of covariance matrix.
			Then, a quaternion multiplication is perfromed as the normalized weights are
			multiplied by the quaternion scaling factor gamma.
		kernel_initializer: Initializer for the quaternion `kernel` weights matrix.
			By default it is 'quaternion'. The 'quaternion_independent' 
			and the usual initializers could also be used.
			(see keras.initializers and init.py).
		bias_initializer: Initializer for the bias vector
			(see keras.initializers).
		kernel_regularizer: Regularizer function applied to
			the `kernel` weights matrix
			(see keras.regularizers).
		bias_regularizer: Regularizer function applied to the bias vector
			(see keras.regularizers).
		activity_regularizer: Regularizer function applied to
			the output of the layer (its "activation").
			(see keras.regularizers).
		kernel_constraint: Constraint function applied to the kernel matrix
			(see keras.constraints).
		bias_constraint: Constraint function applied to the bias vector
			(see keras.constraints).
		spectral_parametrization: Whether or not to use a spectral
			parametrization of the parameters.
	# Input shape
		3D tensor with shape: `(batch_size, steps, input_dim)`
	# Output shape
		3D tensor with shape: `(batch_size, new_steps, 2 x filters)`
		`steps` value might have changed due to padding or strides.
	"""

	def __init__(self, filters,
				 kernel_size,
				 strides=1,
				 padding='valid',
				 dilation_rate=1,
				 activation=None,
				 use_bias=True,
				 kernel_initializer='quaternion',
				 bias_initializer='zeros',
				 kernel_regularizer=None,
				 bias_regularizer=None,
				 activity_regularizer=None,
				 kernel_constraint=None,
				 bias_constraint=None,
				 seed=None,
				 init_criterion='he',
				 spectral_parametrization=False,
				 **kwargs):
		super(QuaternionConv1D, self).__init__(
			rank=1,
			filters=filters,
			kernel_size=kernel_size,
			strides=strides,
			padding=padding,
			data_format='channels_last',
			dilation_rate=dilation_rate,
			activation=activation,
			use_bias=use_bias,
			kernel_initializer=kernel_initializer,
			bias_initializer=bias_initializer,
			kernel_regularizer=kernel_regularizer,
			bias_regularizer=bias_regularizer,
			activity_regularizer=activity_regularizer,
			kernel_constraint=kernel_constraint,
			bias_constraint=bias_constraint,
			init_criterion=init_criterion,
			spectral_parametrization=spectral_parametrization,
			**kwargs)

	def get_config(self):
		config = super(QuaternionConv1D, self).get_config()
		config.pop('rank')
		config.pop('data_format')
		return config


class QuaternionConv2D(QuaternionConv):
	"""2D Quaternion convolution layer (e.g. spatial convolution over images).
	This layer creates a quaternion convolution kernel that is convolved
	with a quaternion input layer to produce a quaternion output tensor. If `use_bias` 
	is True, a quaternion bias vector is created and added to the outputs.
	Finally, if `activation` is not `None`, it is applied to both the
	real and imaginary parts of the output.
	When using this layer as the first layer in a model,
	provide the keyword argument `input_shape`
	(tuple of integers, does not include the sample axis),
	e.g. `input_shape=(128, 128, 3)` for 128x128 RGB pictures
	in `data_format="channels_last"`.
	# Arguments
		filters: Integer, the dimensionality of the quaternion output space
			(i.e, the number quaternion feature maps in the convolution).
			The total effective number of filters or feature maps is 2 x filters.
		kernel_size: An integer or tuple/list of 2 integers, specifying the
			width and height of the 2D convolution window.
			Can be a single integer to specify the same value for
			all spatial dimensions.
		strides: An integer or tuple/list of 2 integers,
			specifying the strides of the convolution along the width and height.
			Can be a single integer to specify the same value for
			all spatial dimensions.
			Specifying any stride value != 1 is incompatible with specifying
			any `dilation_rate` value != 1.
		padding: one of `"valid"` or `"same"` (case-insensitive).
		data_format: A string,
			one of `channels_last` (default) or `channels_first`.
			The ordering of the dimensions in the inputs.
			`channels_last` corresponds to inputs with shape
			`(batch, height, width, channels)` while `channels_first`
			corresponds to inputs with shape
			`(batch, channels, height, width)`.
			It defaults to the `image_data_format` value found in your
			Keras config file at `~/.keras/keras.json`.
			If you never set it, then it will be "channels_last".
		dilation_rate: an integer or tuple/list of 2 integers, specifying
			the dilation rate to use for dilated convolution.
			Can be a single integer to specify the same value for
			all spatial dimensions.
			Currently, specifying any `dilation_rate` value != 1 is
			incompatible with specifying any stride value != 1.
		activation: Activation function to use
			(see keras.activations).
			If you don't specify anything, no activation is applied
			(ie. "linear" activation: `a(x) = x`).
		use_bias: Boolean, whether the layer uses a bias vector.
		normalize_weight: Boolean, whether the layer normalizes its quaternion
			weights before convolving the quaternion input.
			The quaternion normalization performed is similar to the one
			for the batchnorm. Each of the quaternion kernels are centred and multiplied by
			the inverse square root of covariance matrix.
			Then, a quaternion multiplication is perfromed as the normalized weights are
			multiplied by the quaternion scaling factor gamma.
		kernel_initializer: Initializer for the quaternion `kernel` weights matrix.
			By default it is 'quaternion'. The 'quaternion_independent' 
			and the usual initializers could also be used.
			(see keras.initializers and init.py).
		bias_initializer: Initializer for the bias vector
			(see keras.initializers).
		kernel_regularizer: Regularizer function applied to
			the `kernel` weights matrix
			(see keras.regularizers).
		bias_regularizer: Regularizer function applied to the bias vector
			(see keras.regularizers).
		activity_regularizer: Regularizer function applied to
			the output of the layer (its "activation").
			(see keras.regularizers).
		kernel_constraint: Constraint function applied to the kernel matrix
			(see keras.constraints).
		bias_constraint: Constraint function applied to the bias vector
			(see keras.constraints).
		spectral_parametrization: Whether or not to use a spectral
			parametrization of the parameters.
	# Input shape
		4D tensor with shape:
		`(samples, channels, rows, cols)` if data_format='channels_first'
		or 4D tensor with shape:
		`(samples, rows, cols, channels)` if data_format='channels_last'.
	# Output shape
		4D tensor with shape:
		`(samples, 2 x filters, new_rows, new_cols)` if data_format='channels_first'
		or 4D tensor with shape:
		`(samples, new_rows, new_cols, 2 x filters)` if data_format='channels_last'.
		`rows` and `cols` values might have changed due to padding.
	"""

	def __init__(self, filters,
				 kernel_size,
				 strides=(1, 1),
				 padding='valid',
				 data_format=None,
				 dilation_rate=(1, 1),
				 activation=None,
				 use_bias=True,
				 kernel_initializer='quaternion',
				 bias_initializer='zeros',
				 kernel_regularizer=None,
				 bias_regularizer=None,
				 activity_regularizer=None,
				 kernel_constraint=None,
				 bias_constraint=None,
				 seed=None,
				 init_criterion='he',
				 spectral_parametrization=False,
				 **kwargs):
		super(QuaternionConv2D, self).__init__(
			rank=2,
			filters=filters,
			kernel_size=kernel_size,
			strides=strides,
			padding=padding,
			data_format=data_format,
			dilation_rate=dilation_rate,
			activation=activation,
			use_bias=use_bias,
			kernel_initializer=kernel_initializer,
			bias_initializer=bias_initializer,
			kernel_regularizer=kernel_regularizer,
			bias_regularizer=bias_regularizer,
			activity_regularizer=activity_regularizer,
			kernel_constraint=kernel_constraint,
			bias_constraint=bias_constraint,
			init_criterion=init_criterion,
			spectral_parametrization=spectral_parametrization,
			**kwargs)

	def get_config(self):
		config = super(QuaternionConv2D, self).get_config()
		config.pop('rank')
		return config


class QuaternionConv3D(QuaternionConv):
	"""3D convolution layer (e.g. spatial convolution over volumes).
	This layer creates a quaternion convolution kernel that is convolved
	with a quaternion layer input to produce a quaternion output tensor.
	If `use_bias` is True,
	a quaternion bias vector is created and added to the outputs. Finally, if
	`activation` is not `None`, it is applied to each of the real and imaginary
	parts of the output.
	When using this layer as the first layer in a model,
	provide the keyword argument `input_shape`
	(tuple of integers, does not include the sample axis),
	e.g. `input_shape=(2, 128, 128, 128, 3)` for 128x128x128 volumes
	with 3 channels,
	in `data_format="channels_last"`.
	# Arguments
		filters: Integer, the dimensionality of the quaternion output space
			(i.e, the number quaternion feature maps in the convolution).
			The total effective number of filters or feature maps is 2 x filters.
		kernel_size: An integer or tuple/list of 3 integers, specifying the
			width and height of the 3D convolution window.
			Can be a single integer to specify the same value for
			all spatial dimensions.
		strides: An integer or tuple/list of 3 integers,
			specifying the strides of the convolution along each spatial dimension.
			Can be a single integer to specify the same value for
			all spatial dimensions.
			Specifying any stride value != 1 is incompatible with specifying
			any `dilation_rate` value != 1.
		padding: one of `"valid"` or `"same"` (case-insensitive).
		data_format: A string,
			one of `channels_last` (default) or `channels_first`.
			The ordering of the dimensions in the inputs.
			`channels_last` corresponds to inputs with shape
			`(batch, spatial_dim1, spatial_dim2, spatial_dim3, channels)`
			while `channels_first` corresponds to inputs with shape
			`(batch, channels, spatial_dim1, spatial_dim2, spatial_dim3)`.
			It defaults to the `image_data_format` value found in your
			Keras config file at `~/.keras/keras.json`.
			If you never set it, then it will be "channels_last".
		dilation_rate: an integer or tuple/list of 3 integers, specifying
			the dilation rate to use for dilated convolution.
			Can be a single integer to specify the same value for
			all spatial dimensions.
			Currently, specifying any `dilation_rate` value != 1 is
			incompatible with specifying any stride value != 1.
		activation: Activation function to use
			(see keras.activations).
			If you don't specify anything, no activation is applied
			(ie. "linear" activation: `a(x) = x`).
		use_bias: Boolean, whether the layer uses a bias vector.
		normalize_weight: Boolean, whether the layer normalizes its quaternion
			weights before convolving the quaternion input.
			The quaternion normalization performed is similar to the one
			for the batchnorm. Each of the quaternion kernels are centred and multiplied by
			the inverse square root of covariance matrix.
			Then, a quaternion multiplication is perfromed as the normalized weights are
			multiplied by the quaternion scaling factor gamma.
		kernel_initializer: Initializer for the quaternion `kernel` weights matrix.
			By default it is 'quaternion'. The 'quaternion_independent' 
			and the usual initializers could also be used.
			(see keras.initializers and init.py).
		bias_initializer: Initializer for the bias vector
			(see keras.initializers).
		kernel_regularizer: Regularizer function applied to
			the `kernel` weights matrix
			(see keras.regularizers).
		bias_regularizer: Regularizer function applied to the bias vector
			(see keras.regularizers).
		activity_regularizer: Regularizer function applied to
			the output of the layer (its "activation").
			(see keras.regularizers).
		kernel_constraint: Constraint function applied to the kernel matrix
			(see keras.constraints).
		bias_constraint: Constraint function applied to the bias vector
			(see keras.constraints).
		spectral_parametrization: Whether or not to use a spectral
			parametrization of the parameters.
	# Input shape
		5D tensor with shape:
		`(samples, channels, conv_dim1, conv_dim2, conv_dim3)` if data_format='channels_first'
		or 5D tensor with shape:
		`(samples, conv_dim1, conv_dim2, conv_dim3, channels)` if data_format='channels_last'.
	# Output shape
		5D tensor with shape:
		`(samples, 2 x filters, new_conv_dim1, new_conv_dim2, new_conv_dim3)` if data_format='channels_first'
		or 5D tensor with shape:
		`(samples, new_conv_dim1, new_conv_dim2, new_conv_dim3, 2 x filters)` if data_format='channels_last'.
		`new_conv_dim1`, `new_conv_dim2` and `new_conv_dim3` values might have changed due to padding.
	"""

	def __init__(self, filters,
				 kernel_size,
				 strides=(1, 1, 1),
				 padding='valid',
				 data_format=None,
				 dilation_rate=(1, 1, 1),
				 activation=None,
				 use_bias=True,
				 kernel_initializer='quaternion',
				 bias_initializer='zeros',
				 kernel_regularizer=None,
				 bias_regularizer=None,
				 activity_regularizer=None,
				 kernel_constraint=None,
				 bias_constraint=None,
				 seed=None,
				 init_criterion='he',
				 spectral_parametrization=False,
				 **kwargs):
		super(QuaternionConv3D, self).__init__(
			rank=3,
			filters=filters,
			kernel_size=kernel_size,
			strides=strides,
			padding=padding,
			data_format=data_format,
			dilation_rate=dilation_rate,
			activation=activation,
			use_bias=use_bias,
			kernel_initializer=kernel_initializer,
			bias_initializer=bias_initializer,
			kernel_regularizer=kernel_regularizer,
			bias_regularizer=bias_regularizer,
			activity_regularizer=activity_regularizer,
			kernel_constraint=kernel_constraint,
			bias_constraint=bias_constraint,
			init_criterion=init_criterion,
			spectral_parametrization=spectral_parametrization,
			**kwargs)

	def get_config(self):
		config = super(QuaternionConv3D, self).get_config()
		config.pop('rank')
		return config

def sanitizedInitGet(init):
	if   init in ["sqrt_init"]:
		return sqrt_init
	elif init in ["complex", "complex_independent",
				  "glorot_complex", "he_complex",
				  "quaternion", "quaternion_independent"]:
		return init
	else:
		return initializers.get(init)
def sanitizedInitSer(init):
	if init in [sqrt_init]:
		return "sqrt_init"
	elif init == "complex" or isinstance(init, ComplexInit):
		return "complex"
	elif init == "complex_independent" or isinstance(init, ComplexIndependentFilters):
		return "complex_independent"
	elif init == "quaternion" or isinstance(init, QuaternionInit):
		return "quaternion"
	elif init == "quaternion_independent" or isinstance(init, QuaternionIndependentFilters):
		return "quaternion_independent"
	else:
		return initializers.serialize(init)


# Aliases
QuaternionConvolution1D = QuaternionConv1D
QuaternionConvolution2D = QuaternionConv2D
QuaternionConvolution3D = QuaternionConv3D