GCC Code Coverage Report

Line	Branch	Exec	Source
1			module athena__conv1d_layer
2			!! Module containing implementation of a 1D convolutional layer
3			!!
4			!! This module implements 1D convolution for processing sequential data
5			!! such as time series or text. Applies learnable filters along sequence.
6			!!
7			!! Mathematical operation:
8			!! \[
9			!! y_{i,k}
10			!! =
11			!! \sigma\left(
12			!! \sum_{c=1}^{C_{in}}
13			!! \sum_{m=0}^{K_w-1}
14			!! x_{i+m,\,c}\; w_{m,\,c,\,k}
15			!! + b_k
16			!! \right)
17			!! \]
18			!!
19			!! where:
20			!! - \(i\) is the spatial coordinate in the output
21			!! - \(k\) is the output channel (filter) index
22			!! - \(m\) is the kernel offset
23			!! - \(c\) is the input channel index
24			!! - \(K_w\) is the kernel width
25			!! - \(\sigma\) is the activation function
26			!!
27			!! Shape: \((W, C_{in}) \rightarrow (W', C_{out})\)
28			use coreutils, only: real32, stop_program
29			use athena__base_layer, only: conv_layer_type, base_layer_type
30			use athena__pad1d_layer, only: pad1d_layer_type
31			use athena__misc_types, only: base_actv_type, base_init_type, &
32			onnx_node_type, onnx_initialiser_type, onnx_tensor_type
33			use diffstruc, only: array_type
34			use athena__diffstruc_extd, only: conv1d, add_bias
35			implicit none
36
37
38			private
39
40			public :: conv1d_layer_type
41			public :: read_conv1d_layer
42
43
44			type, extends(conv_layer_type) :: conv1d_layer_type
45			!! Type for 1D convolutional layer with overloaded procedures
46			type(array_type), dimension(2) :: z
47			!! Temporary arrays for forward propagation
48			contains
49			procedure, pass(this) :: set_hyperparams => set_hyperparams_conv1d
50			!! Set hyperparameters for 1D convolutional layer
51			procedure, pass(this) :: read => read_conv1d
52			!! Read 1D convolutional layer from file
53
54			procedure, pass(this) :: forward => forward_conv1d
55			!! Forward propagation derived type handler
56
57			final :: finalise_conv1d
58			!! Finalise 1D convolutional layer
59			end type conv1d_layer_type
60
61			interface conv1d_layer_type
62			!! Interface for setting up the 1D convolutional layer
63			module function layer_setup( &
64			input_shape, &
65			num_filters, kernel_size, stride, dilation, padding, &
66			use_bias, &
67			activation, &
68			kernel_initialiser, bias_initialiser, &
69			verbose ) result(layer)
70			!! Set up the 1D convolutional layer
71			integer, dimension(:), optional, intent(in) :: input_shape
72			!! Input shape
73			integer, optional, intent(in) :: num_filters
74			!! Number of filters
75			integer, dimension(..), optional, intent(in) :: kernel_size
76			!! Kernel size
77			integer, dimension(..), optional, intent(in) :: stride
78			!! Stride
79			integer, dimension(..), optional, intent(in) :: dilation
80			!! Dilation
81			logical, optional, intent(in) :: use_bias
82			!! Use bias
83			class(*), optional, intent(in) :: activation, &
84			kernel_initialiser, bias_initialiser
85			!! Activation function, kernel initialiser, bias initialiser
86			character(*), optional, intent(in) :: padding
87			!! Padding method
88			integer, optional, intent(in) :: verbose
89			!! Verbosity level
90			type(conv1d_layer_type) :: layer
91			!! Instance of the 1D convolutional layer
92			end function layer_setup
93			end interface conv1d_layer_type
94
95
96
97			contains
98
99			!###############################################################################
100		−	subroutine finalise_conv1d(this)
101			!! Finalise 1D convolutional layer
102			implicit none
103
104			! Arguments
105			type(conv1d_layer_type), intent(inout) :: this
106			!! Instance of the 1D convolutional layer
107
108		−	if(allocated(this%dil)) deallocate(this%dil)
109		−	if(allocated(this%knl)) deallocate(this%knl)
110		−	if(allocated(this%stp)) deallocate(this%stp)
111
112		−	if(allocated(this%input_shape)) deallocate(this%input_shape)
113		−	if(allocated(this%output)) deallocate(this%output)
114		−	if(allocated(this%pad_layer)) deallocate(this%pad_layer)
115
116		−	end subroutine finalise_conv1d
117			!###############################################################################
118
119
120			!##############################################################################!
121			! * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * !
122			!##############################################################################!
123
124
125			!###############################################################################
126		−	module function layer_setup( &
127		−	input_shape, &
128			num_filters, kernel_size, stride, dilation, padding, &
129			use_bias, &
130			activation, &
131			kernel_initialiser, bias_initialiser, &
132		−	verbose ) result(layer)
133			!! Set up the 1D convolutional layer
134			use athena__activation, only: activation_setup
135			use athena__initialiser, only: initialiser_setup
136			implicit none
137
138			! Arguments
139			integer, dimension(:), optional, intent(in) :: input_shape
140			!! Input shape
141			integer, optional, intent(in) :: num_filters
142			!! Number of filters
143			integer, dimension(..), optional, intent(in) :: kernel_size
144			!! Kernel size
145			integer, dimension(..), optional, intent(in) :: stride
146			!! Stride
147			integer, dimension(..), optional, intent(in) :: dilation
148			!! Dilation
149			logical, optional, intent(in) :: use_bias
150			!! Use bias
151			class(*), optional, intent(in) :: activation
152			!! Activation function
153			class(*), optional, intent(in) :: kernel_initialiser, bias_initialiser
154			!! Activation function, kernel initialiser, and bias initialiser
155			character(*), optional, intent(in) :: padding
156			!! Padding method
157			integer, optional, intent(in) :: verbose
158			!! Verbosity level
159
160			type(conv1d_layer_type) :: layer
161			!! Instance of the 1D convolutional layer
162
163			! Local variables
164			integer :: verbose_ = 0
165			!! Verbosity level
166			integer :: num_filters_
167			!! Number of filters
168			logical :: use_bias_ = .true.
169			!! Use bias
170			character(len=10) :: activation_function_ = "none"
171			!! Activation function
172			character(len=20) :: padding_
173			!! Padding
174			integer, dimension(1) :: kernel_size_, stride_, dilation_
175			!! Kernel size and stride
176		−	class(base_actv_type), allocatable :: activation_
177			!! Activation function
178		−	class(base_init_type), allocatable :: kernel_initialiser_, bias_initialiser_
179			!! Kernel and bias initialisers
180
181		−	if(present(verbose)) verbose_ = verbose
182
183
184			!---------------------------------------------------------------------------
185			! Set use_bias
186			!---------------------------------------------------------------------------
187		−	if(present(use_bias)) use_bias_ = use_bias
188
189
190			!---------------------------------------------------------------------------
191			! Set activation functions based on input name
192			!---------------------------------------------------------------------------
193		−	if(present(activation))then
194		−	activation_ = activation_setup(activation)
195			else
196		−	activation_ = activation_setup("none")
197			end if
198
199
200			!---------------------------------------------------------------------------
201			! Define weights (kernels) and biases initialisers
202			!---------------------------------------------------------------------------
203		−	if(present(kernel_initialiser))then
204		−	kernel_initialiser_ = initialiser_setup(kernel_initialiser)
205			end if
206		−	if(present(bias_initialiser))then
207		−	bias_initialiser_ = initialiser_setup(bias_initialiser)
208			end if
209
210
211			!---------------------------------------------------------------------------
212			! Set up number of filters
213			!---------------------------------------------------------------------------
214		−	if(present(num_filters))then
215		−	num_filters_ = num_filters
216			else
217		−	num_filters_ = 32
218			end if
219
220
221			!---------------------------------------------------------------------------
222			! Set up kernel size
223			!---------------------------------------------------------------------------
224		−	if(present(kernel_size))then
225		−	select rank(kernel_size)
226			rank(0)
227		−	kernel_size_ = kernel_size
228			rank(1)
229		−	kernel_size_ = kernel_size
230			end select
231			else
232		−	kernel_size_ = 3
233			end if
234
235
236			!---------------------------------------------------------------------------
237			! Set up dilation
238			!---------------------------------------------------------------------------
239		−	if(present(dilation))then
240		−	select rank(dilation)
241			rank(0)
242		−	dilation_ = dilation
243			rank(1)
244		−	dilation_ = dilation(1)
245			end select
246			else
247		−	dilation_ = 1
248			end if
249
250
251			!---------------------------------------------------------------------------
252			! Set up padding name
253			!---------------------------------------------------------------------------
254		−	if(present(padding))then
255		−	padding_ = padding
256			else
257		−	padding_ = "valid"
258			end if
259
260
261			!---------------------------------------------------------------------------
262			! Set up stride
263			!---------------------------------------------------------------------------
264		−	if(present(stride))then
265		−	select rank(stride)
266			rank(0)
267		−	stride_ = stride
268			rank(1)
269		−	stride_ = stride(1)
270			end select
271			else
272		−	stride_ = 1
273			end if
274
275
276			!---------------------------------------------------------------------------
277			! Set hyperparameters
278			!---------------------------------------------------------------------------
279			call layer%set_hyperparams( &
280			num_filters = num_filters_, &
281			kernel_size = kernel_size_, stride = stride_, dilation = dilation_, &
282			padding = padding_, &
283			use_bias = use_bias_, &
284			activation = activation_, &
285			kernel_initialiser = kernel_initialiser_, &
286			bias_initialiser = bias_initialiser_, &
287			verbose = verbose_ &
288		−	)
289
290
291			!---------------------------------------------------------------------------
292			! Initialise layer shape
293			!---------------------------------------------------------------------------
294		−	if(present(input_shape)) call layer%init(input_shape=input_shape)
295
296		−	end function layer_setup
297			!###############################################################################
298
299
300			!###############################################################################
301		−	subroutine set_hyperparams_conv1d( &
302			this, &
303			num_filters, &
304			kernel_size, stride, dilation, &
305			padding, &
306			use_bias, &
307			activation, &
308			kernel_initialiser, bias_initialiser, &
309			verbose &
310			)
311			!! Set hyperparameters for 1D convolutional layer
312			use athena__activation, only: activation_setup
313			use athena__initialiser, only: get_default_initialiser, initialiser_setup
314			use coreutils, only: to_lower
315			implicit none
316
317			! Arguments
318			class(conv1d_layer_type), intent(inout) :: this
319			!! Instance of the 1D convolutional layer
320			integer, intent(in) :: num_filters
321			!! Number of filters
322			integer, dimension(1), intent(in) :: kernel_size, stride, dilation
323			!! Kernel size, stride, dilation
324			character(*), intent(in) :: padding
325			!! Padding
326			logical, intent(in) :: use_bias
327			!! Use bias
328			class(base_actv_type), allocatable, intent(in) :: activation
329			!! Activation function
330			class(base_init_type), allocatable, intent(in) :: &
331			kernel_initialiser, bias_initialiser
332			!! Kernel and bias initialisers
333			integer, optional, intent(in) :: verbose
334			!! Verbosity level
335
336			! Local variables
337			character(len=20) :: padding_
338			character(len=256) :: buffer
339
340		−	this%name = "conv1d"
341		−	this%type = "conv"
342		−	this%input_rank = 2
343		−	this%output_rank = 2
344		−	this%use_bias = use_bias
345		−	if(allocated(this%dil)) deallocate(this%dil)
346		−	if(allocated(this%knl)) deallocate(this%knl)
347		−	if(allocated(this%stp)) deallocate(this%stp)
348			allocate( &
349		−	this%dil(this%input_rank-1), &
350		−	this%knl(this%input_rank-1), &
351		−	this%stp(this%input_rank-1) &
352		−	)
353		−	this%dil = dilation
354		−	this%knl = kernel_size
355		−	this%stp = stride
356		−	this%num_filters = num_filters
357		−	padding_ = trim(adjustl(padding))
358
359		−	select case(trim(adjustl(to_lower(padding_))))
360			case("valid", "none", "")
361			case default
362			this%pad_layer = pad1d_layer_type( &
363		−	padding = [ (this%knl-1)/2 ], &
364			method = padding_ &
365		−	)
366			end select
367		−	if(allocated(this%activation)) deallocate(this%activation)
368		−	if(.not.allocated(activation))then
369		−	this%activation = activation_setup("none")
370			else
371		−	allocate(this%activation, source=activation)
372			end if
373		−	if(allocated(this%kernel_init)) deallocate(this%kernel_init)
374		−	if(.not.allocated(kernel_initialiser))then
375		−	buffer = get_default_initialiser(this%activation%name)
376		−	this%kernel_init = initialiser_setup(buffer)
377			else
378		−	allocate(this%kernel_init, source=kernel_initialiser)
379			end if
380		−	if(allocated(this%bias_init)) deallocate(this%bias_init)
381		−	if(.not.allocated(bias_initialiser))then
382			buffer = get_default_initialiser( &
383			this%activation%name, &
384			is_bias=.true. &
385		−	)
386		−	this%bias_init = initialiser_setup(buffer)
387			else
388		−	allocate(this%bias_init, source=bias_initialiser)
389			end if
390		−	if(present(verbose))then
391		−	if(abs(verbose).gt.0)then
392			write(*,'("CONV1D activation function: ",A)') &
393		−	trim(this%activation%name)
394			write(*,'("CONV1D kernel initialiser: ",A)') &
395		−	trim(this%kernel_init%name)
396			write(*,'("CONV1D bias initialiser: ",A)') &
397		−	trim(this%bias_init%name)
398			end if
399			end if
400
401		−	end subroutine set_hyperparams_conv1d
402			!###############################################################################
403
404
405			!##############################################################################!
406			! * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * !
407			!##############################################################################!
408
409
410			!###############################################################################
411		−	subroutine read_conv1d(this, unit, verbose)
412			!! Read 1D convolutional layer from file
413			use athena__tools_infile, only: assign_val, assign_vec, move
414			use coreutils, only: to_lower, to_upper, icount
415			use athena__activation, only: read_activation
416			use athena__initialiser, only: initialiser_setup
417			implicit none
418
419			! Arguments
420			class(conv1d_layer_type), intent(inout) :: this
421			!! Instance of the 1D convolutional layer
422			integer, intent(in) :: unit
423			!! Unit number
424			integer, optional, intent(in) :: verbose
425			!! Verbosity level
426
427			! Local variables
428			integer :: stat
429			!! Status of read
430			integer :: verbose_ = 0
431			!! Verbosity level
432			integer :: j, k, l, c, itmp1, iline, num_params
433			!! Loop variables and temporary integer
434			integer :: num_filters
435			!! Number of filters
436			logical :: use_bias = .true.
437			!! Whether to use bias
438			character(14) :: kernel_initialiser_name='', bias_initialiser_name=''
439			!! Kernel and bias initialisers
440			character(20) :: padding='', activation_name=''
441			!! Padding and activation function
442		−	class(base_actv_type), allocatable :: activation
443			!! Activation function
444		−	class(base_init_type), allocatable :: kernel_initialiser, bias_initialiser
445			!! Initialisers
446			character(256) :: buffer, tag, err_msg
447			!! Buffer, tag, and error message
448			integer, dimension(1) :: kernel_size, stride, dilation
449			!! Kernel size and stride
450			integer, dimension(2) :: input_shape
451			!! Input shape
452		−	real(real32), allocatable, dimension(:) :: data_list
453			!! Data list
454			integer :: param_line, final_line
455			!! Parameter line number
456
457
458			! Initialise optional arguments
459			!---------------------------------------------------------------------------
460		−	if(present(verbose)) verbose_ = verbose
461
462
463			! Loop over tags in layer card
464			!---------------------------------------------------------------------------
465		−	iline = 0
466		−	param_line = 0
467		−	final_line = 0
468		−	tag_loop: do
469
470			! Check for end of file
471			!------------------------------------------------------------------------
472		−	read(unit,'(A)',iostat=stat) buffer
473		−	if(stat.ne.0)then
474			write(err_msg,'("file encountered error (EoF?) before END ",A)') &
475		−	to_upper(this%name)
476		−	call stop_program(err_msg)
477		−	return
478			end if
479		−	if(trim(adjustl(buffer)).eq."") cycle tag_loop
480
481			! Check for end of layer card
482			!------------------------------------------------------------------------
483		−	if(trim(adjustl(buffer)).eq."END "//to_upper(trim(this%name)))then
484		−	final_line = iline
485		−	backspace(unit)
486		−	exit tag_loop
487			end if
488		−	iline = iline + 1
489
490		−	tag=trim(adjustl(buffer))
491		−	if(scan(buffer,"=").ne.0) tag=trim(tag(:scan(tag,"=")-1))
492
493			! Read parameters from save file
494			!------------------------------------------------------------------------
495		−	select case(trim(tag))
496			case("INPUT_SHAPE")
497		−	call assign_vec(buffer, input_shape, itmp1)
498			case("NUM_FILTERS")
499		−	call assign_val(buffer, num_filters, itmp1)
500			case("KERNEL_SIZE")
501		−	call assign_vec(buffer, kernel_size, itmp1)
502			case("STRIDE")
503		−	call assign_vec(buffer, stride, itmp1)
504			case("DILATION")
505		−	call assign_vec(buffer, dilation, itmp1)
506			case("USE_BIAS")
507		−	call assign_val(buffer, use_bias, itmp1)
508			case("PADDING")
509		−	call assign_val(buffer, padding, itmp1)
510		−	padding = to_lower(padding)
511			case("ACTIVATION")
512		−	iline = iline - 1
513		−	backspace(unit)
514		−	activation = read_activation(unit, iline)
515			case("KERNEL_INITIALISER", "KERNEL_INIT", "KERNEL_INITIALIZER")
516		−	call assign_val(buffer, kernel_initialiser_name, itmp1)
517			case("BIAS_INITIALISER", "BIAS_INIT", "BIAS_INITIALIZER")
518		−	call assign_val(buffer, bias_initialiser_name, itmp1)
519			case("WEIGHTS")
520		−	kernel_initialiser_name = 'zeros'
521		−	bias_initialiser_name = 'zeros'
522		−	param_line = iline
523			case default
524			! Don't look for "e" due to scientific notation of numbers
525			! ... i.e. exponent (E+00)
526		−	if(scan(to_lower(trim(adjustl(buffer))),&
527			'abcdfghijklmnopqrstuvwxyz').eq.0)then
528		−	cycle tag_loop
529		−	elseif(tag(:3).eq.'END')then
530		−	cycle tag_loop
531			end if
532			write(err_msg,'("Unrecognised line in input file: ",A)') &
533		−	trim(adjustl(buffer))
534		−	call stop_program(err_msg)
535		−	return
536			end select
537			end do tag_loop
538		−	kernel_initialiser = initialiser_setup(kernel_initialiser_name)
539		−	bias_initialiser = initialiser_setup(bias_initialiser_name)
540
541
542			! Set hyperparameters and initialise layer
543			!---------------------------------------------------------------------------
544			call this%set_hyperparams( &
545			num_filters = num_filters, &
546			kernel_size = kernel_size, stride = stride, dilation = dilation, &
547			padding = padding, &
548			use_bias = use_bias, &
549			activation = activation, &
550			kernel_initialiser = kernel_initialiser, &
551			bias_initialiser = bias_initialiser, &
552			verbose = verbose_ &
553		−	)
554		−	call this%init(input_shape = input_shape)
555
556
557			! Check if WEIGHTS card was found
558			!---------------------------------------------------------------------------
559		−	if(param_line.eq.0)then
560		−	write(0,*) "WARNING: WEIGHTS card in "//to_upper(trim(this%name))//" not found"
561			else
562		−	call move(unit, param_line - iline, iostat=stat)
563		−	num_params = product(this%knl) * input_shape(2) * num_filters
564		−	allocate(data_list(num_params), source=0._real32)
565		−	c = 1
566		−	k = 1
567		−	data_concat_loop1: do while(c.le.num_params)
568		−	read(unit,'(A)',iostat=stat) buffer
569		−	if(stat.ne.0) exit data_concat_loop1
570		−	k = icount(buffer)
571		−	read(buffer,*,iostat=stat) (data_list(j),j=c,c+k-1)
572		−	c = c + k
573			end do data_concat_loop1
574		−	this%params(1)%val(:,1) = data_list
575		−	deallocate(data_list)
576		−	allocate(data_list(num_filters), source=0._real32)
577		−	c = 1
578		−	k = 1
579		−	data_concat_loop2: do while(c.le.num_filters)
580		−	read(unit,'(A)',iostat=stat) buffer
581		−	if(stat.ne.0) exit data_concat_loop2
582		−	k = icount(buffer)
583		−	read(buffer,*,iostat=stat) (data_list(j),j=c,c+k-1)
584		−	c = c + k
585			end do data_concat_loop2
586		−	this%params(2)%val(:,1) = data_list
587		−	deallocate(data_list)
588
589			! Check for end of weights card
590			!------------------------------------------------------------------------
591		−	read(unit,'(A)') buffer
592		−	if(trim(adjustl(buffer)).ne."END WEIGHTS")then
593		−	write(0,*) trim(adjustl(buffer))
594		−	call stop_program("END WEIGHTS not where expected")
595		−	return
596			end if
597			end if
598
599
600			! Check for end of layer card
601			!---------------------------------------------------------------------------
602		−	call move(unit, final_line - iline, iostat=stat)
603		−	read(unit,'(A)') buffer
604		−	if(trim(adjustl(buffer)).ne."END "//to_upper(trim(this%name)))then
605		−	write(0,*) trim(adjustl(buffer))
606		−	write(err_msg,'("END ",A," not where expected")') to_upper(this%name)
607		−	call stop_program(err_msg)
608		−	return
609			end if
610
611		−	end subroutine read_conv1d
612			!###############################################################################
613
614
615			!###############################################################################
616		−	function read_conv1d_layer(unit, verbose) result(layer)
617			!! Read 1D convolutional layer from file and return layer
618			implicit none
619
620			! Arguments
621			integer, intent(in) :: unit
622			!! Unit number
623			integer, optional, intent(in) :: verbose
624			!! Verbosity level
625			class(base_layer_type), allocatable :: layer
626			!! Instance of the 1D convolutional layer
627
628			! Local variables
629			integer :: verbose_ = 0
630			!! Verbosity level
631
632		−	if(present(verbose)) verbose_ = verbose
633		−	allocate(layer, source=conv1d_layer_type())
634		−	call layer%read(unit, verbose=verbose_)
635
636		−	end function read_conv1d_layer
637			!###############################################################################
638
639
640			!###############################################################################
641		−	subroutine build_from_onnx_conv1d( &
642		−	this, node, initialisers, value_info, verbose &
643			)
644			!! Read ONNX attributes for 1D convolutional layer
645			use athena__activation, only: activation_setup
646			use athena__initialiser_data, only: data_init_type
647			implicit none
648
649			! Arguments
650			class(conv1d_layer_type), intent(inout) :: this
651			!! Instance of the 1D convolutional layer
652			type(onnx_node_type), intent(in) :: node
653			!! ONNX node information
654			type(onnx_initialiser_type), dimension(:), intent(in) :: initialisers
655			!! ONNX initialiser information
656			type(onnx_tensor_type), dimension(:), intent(in) :: value_info
657			!! ONNX value info information
658			integer, intent(in) :: verbose
659			!! Verbosity level
660
661			! Local variables
662			integer :: i, weight_idx, bias_idx
663			!! Loop index and temporary integer
664			integer :: num_filters
665			!! Number of filters
666			logical :: use_bias = .true.
667			!! Whether to use bias
668			integer, dimension(1) :: padding, stride, kernel_size, dilation
669			!! Padding, stride, kernel size, and dilation
670		−	integer, dimension(:), allocatable :: dims
671			!! Dimensions
672			character(256) :: val
673			!! Attribute value
674		−	class(base_actv_type), allocatable :: activation
675			!! Activation function
676		−	class(base_init_type), allocatable :: kernel_initialiser, bias_initialiser
677
678			! Set default values
679		−	padding = 0
680		−	stride = 1
681		−	kernel_size = 3
682		−	dilation = 1
683
684		−	do i = 1, size(node%attributes)
685		−	val = node%attributes(i)%val
686		−	select case(trim(adjustl(node%attributes(i)%name)))
687			case("pads")
688		−	read(val,*) padding
689			case("strides")
690		−	read(val,*) stride
691			case("kernel_shape")
692		−	read(val,*) kernel_size
693			case("dilations")
694		−	read(val,*) dilation
695			case default
696			! Do nothing
697		−	write(0,*) "WARNING: Unrecognised attribute in ONNX CONV1D layer: ", &
698		−	trim(adjustl(node%attributes(i)%name))
699			end select
700			end do
701
702		−	weight_idx = -1
703		−	bias_idx = -1
704		−	allocate(dims(0))
705		−	if(size(initialisers).lt.1)then
706		−	call stop_program("ONNX CONV1D layer requires at least 1 initialiser")
707		−	return
708			else
709			! check which initialiser has weights and which has biases
710		−	do i = 1, size(initialisers)
711		−	if(allocated(initialisers(i)%dims))then
712		−	dims = [ dims, product(initialisers(i)%dims) ]
713			end if
714			end do
715			end if
716
717		−	select case(size(dims))
718			case(1)
719		−	if(mod(dims(1), kernel_size(1)).eq.0)then
720		−	weight_idx = 1
721			else
722			call stop_program("ONNX CONV1D layer initialiser dimensions do not &
723		−	&match kernel size")
724		−	return
725			end if
726		−	use_bias = .false.
727			case(2)
728			! check which is weight and which is bias
729		−	if(mod(dims(1), kernel_size(1)).eq.0 .and. &
730		−	dims(1)/kernel_size(1).eq.dims(2))then
731		−	weight_idx = 1
732		−	bias_idx = 2
733		−	elseif(mod(dims(2), kernel_size(1)).eq.0 .and. &
734		−	dims(2)/kernel_size(1).eq.dims(1))then
735		−	weight_idx = 2
736		−	bias_idx = 1
737			else
738			call stop_program("ONNX CONV1D layer initialiser dimensions do not &
739		−	&match kernel size")
740		−	return
741			end if
742			case default
743		−	call stop_program("ONNX CONV1D layer number of initialisers not supported")
744		−	return
745			end select
746
747		−	num_filters = dims(weight_idx) / kernel_size(1)
748		−	if(num_filters .ne. value_info(1)%dims(2))then
749			call stop_program("ONNX CONV1D layer number of filters does not match &
750		−	&value info")
751		−	return
752			end if
753
754		−	kernel_initialiser = data_init_type( data = initialisers(weight_idx)%data )
755		−	if(use_bias)then
756		−	bias_initialiser = data_init_type( data = initialisers(bias_idx)%data )
757			end if
758
759		−	activation = activation_setup("none")
760			call this%set_hyperparams( &
761			num_filters = num_filters, &
762			kernel_size = kernel_size, stride = stride, &
763			dilation = dilation, &
764			padding = "valid", &
765			use_bias = use_bias, &
766			activation = activation, &
767			verbose = verbose, &
768			kernel_initialiser = kernel_initialiser, &
769			bias_initialiser = bias_initialiser &
770		−	)
771
772		−	end subroutine build_from_onnx_conv1d
773			!###############################################################################
774
775
776			!##############################################################################!
777			! * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * !
778			!##############################################################################!
779
780
781			!###############################################################################
782		−	subroutine forward_conv1d(this, input)
783			!! Forward propagation
784			implicit none
785
786			! Arguments
787			class(conv1d_layer_type), intent(inout) :: this
788			!! Instance of the 1D convolutional layer
789			class(array_type), dimension(:,:), intent(in) :: input
790			!! Input values
791
792			! Local variables
793			type(array_type), pointer :: ptr
794			!! Pointer array
795
796
797			! Generate outputs from weights, biases, and inputs
798			!---------------------------------------------------------------------------
799		−	select case(allocated(this%pad_layer))
800			case(.true.)
801		−	call this%pad_layer%forward(input)
802		−	ptr => conv1d(this%pad_layer%output(1,1), this%params(1), &
803		−	this%stp(1), this%dil(1) &
804		−	)
805			case default
806		−	ptr => conv1d(input(1,1), this%params(1), this%stp(1), this%dil(1))
807			end select
808		−	ptr => add_bias(ptr, this%params(2), dim=2, dim_act_on_shape=.true.)
809
810			! Apply activation function to activation
811			!---------------------------------------------------------------------------
812		−	if(trim(this%activation%name) .eq. "none") then
813		−	call this%output(1,1)%assign_and_deallocate_source(ptr)
814			else
815		−	ptr => this%activation%apply(ptr)
816		−	call this%output(1,1)%assign_and_deallocate_source(ptr)
817			end if
818		−	this%output(1,1)%is_temporary = .false.
819
820		−	end subroutine forward_conv1d
821			!###############################################################################
822
823		−	end module athena__conv1d_layer
824