@@ -284,77 +284,187 @@ final class TensorAutoDiffTests: XCTestCase {
284284 XCTAssertEqual ( varianceGradAlongAxes ( input) , expected)
285285 }
286286
287- func testMin( ) {
288- // The expected gradient values were computed using the following TensorFlow 2.0 Beta1
289- // Python code with respective `a` and `b` tensors:
287+ func testMax( ) {
288+ // Expected gradient values were computed using the following TensorFlow Python code:
290289 // ```
290+ // import tensorflow as tf
291291 // with tf.GradientTape() as t:
292292 // t.watch([a, b])
293- // y = tf.math. reduce_sum(tf.minimum (a, b))
293+ // y = tf.reduce_sum(tf.maximum (a, b))
294294 // print(t.gradient(y, [a, b]))
295295 // ```
296296 do {
297297 let a = Tensor < Float > ( [ 4 , 5 , 3 ] )
298298 let b = Tensor < Float > ( [ 4 , 2 , 6 ] )
299- let computedGradient1 = gradient ( at: a, b) { a, b in min ( a, b) . sum ( ) }
300- let expectedGradient1 : ( Tensor < Float > , Tensor < Float > ) = (
301- [ 1.0 , 0.0 , 1.0 ] , [ 0.0 , 1.0 , 0.0 ] )
299+ let computedGradient1 = gradient ( at: a, b) { a, b in max ( a, b) . sum ( ) }
300+ let expectedGradient1 : ( Tensor < Float > , Tensor < Float > ) = ( [ 1 , 1 , 0 ] , [ 0 , 0 , 1 ] )
302301 XCTAssertEqual ( computedGradient1. 0 , expectedGradient1. 0 )
303302 XCTAssertEqual ( computedGradient1. 1 , expectedGradient1. 1 )
304303
305- let computedGradient2 = gradient ( at: a, b) { a, b in min ( b, a) . sum ( ) }
306- let expectedGradient2 : ( Tensor < Float > , Tensor < Float > ) = (
307- [ 0.0 , 0.0 , 1.0 ] , [ 1.0 , 1.0 , 0.0 ] )
304+ let computedGradient2 = gradient ( at: a, b) { a, b in max ( b, a) . sum ( ) }
305+ let expectedGradient2 : ( Tensor < Float > , Tensor < Float > ) = ( [ 0 , 1 , 0 ] , [ 1 , 0 , 1 ] )
308306 XCTAssertEqual ( computedGradient2. 0 , expectedGradient2. 0 )
309307 XCTAssertEqual ( computedGradient2. 1 , expectedGradient2. 1 )
310308 }
311-
312309 do {
313- let a = Tensor < Float > ( [ [ 3.0 , - 2.0 ] , [ 0.3 , 10.0 ] ] )
314- let b = Tensor < Float > ( [ 9.0 , - 3.0 ] )
315- let computedGradient = gradient ( at: a, b) { a, b in min ( a, b) . sum ( ) }
316- let expectedGradient : ( Tensor < Float > , Tensor < Float > ) = (
317- [ [ 1.0 , 0.0 ] , [ 1.0 , 0.0 ] ] , [ 0.0 , 2.0 ] )
310+ let a = Tensor < Float > ( [ [ 3 , - 2 ] , [ 0.3 , 10 ] ] )
311+ let b = Tensor < Float > ( [ 9 , - 3 ] )
312+ let computedGradient = gradient ( at: a, b) { a, b in max ( a, b) . sum ( ) }
313+ let expectedGradient : ( Tensor < Float > , Tensor < Float > ) = ( [ [ 0 , 1 ] , [ 0 , 1 ] ] , [ 2 , 0 ] )
318314 XCTAssertEqual ( computedGradient. 0 , expectedGradient. 0 )
319315 XCTAssertEqual ( computedGradient. 1 , expectedGradient. 1 )
320316 }
321317 }
322318
323- func testMax( ) {
324- // The expected gradient values were computed using the following TensorFlow 2.0 Beta1
325- // Python code with respective `a` and `b` tensors:
319+ func testMin( ) {
320+ // Expected gradient values were computed using the following TensorFlow Python code:
326321 // ```
322+ // import tensorflow as tf
327323 // with tf.GradientTape() as t:
328324 // t.watch([a, b])
329- // y = tf.math. reduce_sum(tf.maximum (a, b))
325+ // y = tf.reduce_sum(tf.minimum (a, b))
330326 // print(t.gradient(y, [a, b]))
331327 // ```
332328 do {
333329 let a = Tensor < Float > ( [ 4 , 5 , 3 ] )
334330 let b = Tensor < Float > ( [ 4 , 2 , 6 ] )
335- let computedGradient1 = gradient ( at: a, b) { a, b in max ( a, b) . sum ( ) }
336- let expectedGradient1 : ( Tensor < Float > , Tensor < Float > ) = (
337- [ 1.0 , 1.0 , 0.0 ] , [ 0.0 , 0.0 , 1.0 ] )
331+ let computedGradient1 = gradient ( at: a, b) { a, b in min ( a, b) . sum ( ) }
332+ let expectedGradient1 : ( Tensor < Float > , Tensor < Float > ) = ( [ 1 , 0 , 1 ] , [ 0 , 1 , 0 ] )
338333 XCTAssertEqual ( computedGradient1. 0 , expectedGradient1. 0 )
339334 XCTAssertEqual ( computedGradient1. 1 , expectedGradient1. 1 )
340335
341- let computedGradient2 = gradient ( at: a, b) { a, b in max ( b, a) . sum ( ) }
342- let expectedGradient2 : ( Tensor < Float > , Tensor < Float > ) = (
343- [ 0.0 , 1.0 , 0.0 ] , [ 1.0 , 0.0 , 1.0 ] )
336+ let computedGradient2 = gradient ( at: a, b) { a, b in min ( b, a) . sum ( ) }
337+ let expectedGradient2 : ( Tensor < Float > , Tensor < Float > ) = ( [ 0 , 0 , 1 ] , [ 1 , 1 , 0 ] )
344338 XCTAssertEqual ( computedGradient2. 0 , expectedGradient2. 0 )
345339 XCTAssertEqual ( computedGradient2. 1 , expectedGradient2. 1 )
346340 }
341+
347342 do {
348- let a = Tensor < Float > ( [ [ 3.0 , - 2.0 ] , [ 0.3 , 10.0 ] ] )
349- let b = Tensor < Float > ( [ 9.0 , - 3.0 ] )
350- let computedGradient = gradient ( at: a, b) { a, b in max ( a, b) . sum ( ) }
351- let expectedGradient : ( Tensor < Float > , Tensor < Float > ) = (
352- [ [ 0.0 , 1.0 ] , [ 0.0 , 1.0 ] ] , [ 2.0 , 0.0 ] )
343+ let a = Tensor < Float > ( [ [ 3 , - 2 ] , [ 0.3 , 10 ] ] )
344+ let b = Tensor < Float > ( [ 9 , - 3 ] )
345+ let computedGradient = gradient ( at: a, b) { a, b in min ( a, b) . sum ( ) }
346+ let expectedGradient : ( Tensor < Float > , Tensor < Float > ) = ( [ [ 1 , 0 ] , [ 1 , 0 ] ] , [ 0 , 2 ] )
353347 XCTAssertEqual ( computedGradient. 0 , expectedGradient. 0 )
354348 XCTAssertEqual ( computedGradient. 1 , expectedGradient. 1 )
355349 }
356350 }
357351
352+ func testMaxAlongAxes( ) {
353+ // Expected gradient values were computed using the following TensorFlow Python code:
354+ // ```
355+ // import tensorflow as tf
356+ // x = tf.constant(range(6), shape=(2, 3), dtype=float)
357+ // with tf.GradientTape() as t:
358+ // t.watch(x)
359+ // y = tf.reduce_sum(tf.reduce_max(x, axis=0, keepdims=True))
360+ // print(t.gradient(y, x))
361+ // ```
362+ func maxAlongAxesSum( _ x: Tensor < Float > ) -> Tensor < Float > {
363+ x. max ( alongAxes: 0 ) . sum ( )
364+ }
365+ do {
366+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 3 , 4 , 5 ] ]
367+ let ( value, computedGradient) = valueWithGradient ( at: x, in: maxAlongAxesSum)
368+ XCTAssertEqual ( value, maxAlongAxesSum ( x) )
369+ let expectedGradient : Tensor < Float > = [ [ 0 , 0 , 0 ] , [ 1 , 1 , 1 ] ]
370+ XCTAssertEqual ( computedGradient, expectedGradient)
371+ }
372+ do {
373+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 2 , 1 , 0 ] ]
374+ let ( value, computedGradient) = valueWithGradient ( at: x, in: maxAlongAxesSum)
375+ XCTAssertEqual ( value, maxAlongAxesSum ( x) )
376+ let expectedGradient : Tensor < Float > = [ [ 0 , 0.5 , 1 ] , [ 1 , 0.5 , 0 ] ]
377+ XCTAssertEqual ( computedGradient, expectedGradient)
378+ }
379+ }
380+
381+ func testMinAlongAxes( ) {
382+ // Expected gradient values were computed using the following TensorFlow Python code:
383+ // ```
384+ // import tensorflow as tf
385+ // x = tf.constant(range(6), shape=(2, 3), dtype=float)
386+ // with tf.GradientTape() as t:
387+ // t.watch(x)
388+ // y = tf.reduce_sum(tf.reduce_min(x, axis=0, keepdims=True))
389+ // print(t.gradient(y, x))
390+ // ```
391+ func minAlongAxesSum( _ x: Tensor < Float > ) -> Tensor < Float > {
392+ x. min ( alongAxes: 0 ) . sum ( )
393+ }
394+ do {
395+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 3 , 4 , 5 ] ]
396+ let ( value, computedGradient) = valueWithGradient ( at: x, in: minAlongAxesSum)
397+ XCTAssertEqual ( value, minAlongAxesSum ( x) )
398+ let expectedGradient : Tensor < Float > = [ [ 1 , 1 , 1 ] , [ 0 , 0 , 0 ] ]
399+ XCTAssertEqual ( computedGradient, expectedGradient)
400+ }
401+ do {
402+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 2 , 1 , 0 ] ]
403+ let ( value, computedGradient) = valueWithGradient ( at: x, in: minAlongAxesSum)
404+ XCTAssertEqual ( value, minAlongAxesSum ( x) )
405+ let expectedGradient : Tensor < Float > = [ [ 1 , 0.5 , 0 ] , [ 0 , 0.5 , 1 ] ]
406+ XCTAssertEqual ( computedGradient, expectedGradient)
407+ }
408+ }
409+
410+ func testMaxSqueezingAxes( ) {
411+ // Expected gradient values were computed using the following TensorFlow Python code:
412+ // ```
413+ // import tensorflow as tf
414+ // x = tf.constant(range(6), shape=(2, 3), dtype=float)
415+ // with tf.GradientTape() as t:
416+ // t.watch(x)
417+ // y = tf.reduce_sum(tf.reduce_max(x, axis=0, keepdims=False))
418+ // print(t.gradient(y, x))
419+ // ```
420+ func maxSqueezingAxesSum( _ x: Tensor < Float > ) -> Tensor < Float > {
421+ x. max ( squeezingAxes: 0 ) . sum ( )
422+ }
423+ do {
424+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 3 , 4 , 5 ] ]
425+ let ( value, computedGradient) = valueWithGradient ( at: x, in: maxSqueezingAxesSum)
426+ XCTAssertEqual ( value, maxSqueezingAxesSum ( x) )
427+ let expectedGradient : Tensor < Float > = [ [ 0 , 0 , 0 ] , [ 1 , 1 , 1 ] ]
428+ XCTAssertEqual ( computedGradient, expectedGradient)
429+ }
430+ do {
431+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 2 , 1 , 0 ] ]
432+ let ( value, computedGradient) = valueWithGradient ( at: x, in: maxSqueezingAxesSum)
433+ XCTAssertEqual ( value, maxSqueezingAxesSum ( x) )
434+ let expectedGradient : Tensor < Float > = [ [ 0 , 0.5 , 1 ] , [ 1 , 0.5 , 0 ] ]
435+ XCTAssertEqual ( computedGradient, expectedGradient)
436+ }
437+ }
438+
439+ func testMinSqueezingAxes( ) {
440+ // Expected gradient values were computed using the following TensorFlow Python code:
441+ // ```
442+ // import tensorflow as tf
443+ // x = tf.constant(range(6), shape=(2, 3), dtype=float)
444+ // with tf.GradientTape() as t:
445+ // t.watch(x)
446+ // y = tf.reduce_sum(tf.reduce_min(x, axis=0, keepdims=False))
447+ // print(t.gradient(y, x))
448+ // ```
449+ func minSqueezingAxesSum( _ x: Tensor < Float > ) -> Tensor < Float > {
450+ x. min ( squeezingAxes: 0 ) . sum ( )
451+ }
452+ do {
453+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 3 , 4 , 5 ] ]
454+ let ( value, computedGradient) = valueWithGradient ( at: x, in: minSqueezingAxesSum)
455+ XCTAssertEqual ( value, minSqueezingAxesSum ( x) )
456+ let expectedGradient : Tensor < Float > = [ [ 1 , 1 , 1 ] , [ 0 , 0 , 0 ] ]
457+ XCTAssertEqual ( computedGradient, expectedGradient)
458+ }
459+ do {
460+ let x : Tensor < Float > = [ [ 0 , 1 , 2 ] , [ 2 , 1 , 0 ] ]
461+ let ( value, computedGradient) = valueWithGradient ( at: x, in: minSqueezingAxesSum)
462+ XCTAssertEqual ( value, minSqueezingAxesSum ( x) )
463+ let expectedGradient : Tensor < Float > = [ [ 1 , 0.5 , 0 ] , [ 0 , 0.5 , 1 ] ]
464+ XCTAssertEqual ( computedGradient, expectedGradient)
465+ }
466+ }
467+
358468 func testTensorInitStacking( ) {
359469 let a1 = Tensor < Float > ( [ 1 , 2 , 3 , 4 , 5 ] )
360470 let b1 = Tensor < Float > ( [ 6 , 7 , 8 , 9 , 10 ] )
0 commit comments