Cost after iteration 0 | 389 | #0.6930497356599888 | 390 | #
Cost after iteration 100 | 393 | #0.6464320953428849 | 394 | #
... | 397 | #... | 398 | #
Cost after iteration 2499 | 401 | #0.04421498215868956 | 402 | #
Accuracy | 419 | #0.9999999999999998 | 420 | #
Accuracy | 434 | #0.72 | 435 | #
Cost after iteration 0 | 579 | #0.771749 | 580 | #
Cost after iteration 100 | 583 | #0.672053 | 584 | #
... | 587 | #... | 588 | #
Cost after iteration 2499 | 591 | #0.088439 | 592 | #
606 | # Train Accuracy 607 | # | 608 | #609 | # 0.985645933014 610 | # | 611 | #
Test Accuracy | 625 | #0.8 | 626 | #
Your backward propagation works perfectly fine! | 271 | #difference = 7.814075313343006e-11 | 272 | #
There is a mistake in the backward propagation! | 525 | #difference = 0.2850931567761623 | 526 | #
494 | # Model 495 | # | 496 | #497 | # Train accuracy 498 | # | 499 | #500 | # Problem/Comment 501 | # | 502 | #504 | # 3-layer NN with zeros initialization 505 | # | 506 | #507 | # 50% 508 | # | 509 | #510 | # fails to break symmetry 511 | # | 512 | #
514 | # 3-layer NN with large random initialization 515 | # | 516 | #517 | # 83% 518 | # | 519 | #520 | # too large weights 521 | # | 522 | #
525 | # 3-layer NN with He initialization 526 | # | 527 | #528 | # 99% 529 | # | 530 | #531 | # recommended method 532 | # | 533 | #
471 | # cosine similarity between receptionist and g, before neutralizing: : 472 | # | 473 | #474 | # 0.3307794175059374 475 | # | 476 | #
479 | # cosine similarity between receptionist and g_unit, after neutralizing : 480 | # | 481 | #482 | # 3.5723165491646677e-17 483 | # |
596 | # cosine_similarity(word_to_vec_map["man"], gender) = 597 | # | 598 | #599 | # -0.117110957653 600 | # | 601 | #
604 | # cosine_similarity(word_to_vec_map["woman"], gender) = 605 | # | 606 | #607 | # 0.356666188463 608 | # | 609 | #
616 | # cosine_similarity(e1, gender) = 617 | # | 618 | #619 | # -0.058578740443554995 620 | # | 621 | #
624 | # cosine_similarity(e2, gender) = 625 | # | 626 | #627 | # 0.058578740443555 628 | # | 629 | #