Skip to content

Commit fb90469

Browse files
committed
up
1 parent b9617f8 commit fb90469

2 files changed

Lines changed: 256 additions & 62 deletions

File tree

‎p3_lab_gradient_descent.ipynb‎

Lines changed: 245 additions & 60 deletions
Large diffs are not rendered by default.

‎p3_theory_gradient_descent.ipynb‎

Lines changed: 11 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -440,9 +440,18 @@
440440
"100 에포크의 경사하강법은 처음보다 훨씬 작은 MSE에 도달했지만,\n",
441441
"아직 `LinearRegression`의 결과와는 차이가 있다.\n",
442442
"\n",
443-
"핵심은 두 방법이 같은 계산을 한다는 뜻이 아니다.\n",
443+
"하지만 충분히 학습한 경사하강법은 `LinearRegression`이 찾은 좋은 직선에 가까워진다.\n",
444+
"실제로 5천 에포크까지 훈련시키면 두 모델의 파라미터가 매우 유사해진다.\n",
444445
"\n",
445-
"> **`fit()`은 데이터에 잘 맞는 파라미터를 찾는 과정이고, 경사하강법은 그 파라미터를 반복적인 수정으로 찾아가는 방법이다.**"
446+
"| | 경사하강법, 5000 epoch | `LinearRegression` |\n",
447+
"|---|---:|---:|\n",
448+
"| $\\theta_0$ | 0.795961 | 3.839382 |\n",
449+
"| $\\theta_1$ | 0.660634 | 0.650545 |\n",
450+
"| MSE | 0.650356 | 0.175290 |\n",
451+
"\n",
452+
"**경사하강법 핵심**\n",
453+
"\n",
454+
"> **`fit()`은 결국 데이터에 잘 맞는 파라미터를 찾는 과정이고, 경사하강법은 그 파라미터를 반복적인 수정으로 찾아가는 방법이다.**"
446455
]
447456
},
448457
{

0 commit comments

Comments
 (0)