Skip to content

Instantly share code, notes, and snippets.

@dsaint31x

dsaint31x/dl_classification_loss_activation.ipynb

Created May 23, 2024 12:50
Show Gist options
  • Save dsaint31x/e631a1584f04703d428e3c034e6c63b1 to your computer and use it in GitHub Desktop.
Save dsaint31x/e631a1584f04703d428e3c034e6c63b1 to your computer and use it in GitHub Desktop.
Download ZIP
dl_classification_loss_activation.ipynb
Display the source blob
Display the rendered blob
Raw
dl_classification_loss_activation.ipynb
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"provenance": [],
"authorship_tag": "ABX9TyNfWqHsTiUaC9SiATfQckFu",
"include_colab_link": true
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"language_info": {
"name": "python"
}
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
},
"source": [
"<a href=\"https://colab.research.google.com/gist/dsaint31x/e631a1584f04703d428e3c034e6c63b1/dl_classification_loss_activation.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"cell_type": "code",
"source": [
"import torch\n",
"\n",
"torch.set_printoptions(precision=2)"
],
"metadata": {
"id": "8Dk3a9k9VCe2"
},
"execution_count": 29,
"outputs": []
},
{
"cell_type": "code",
"execution_count": 30,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "lmh9W4ypRv-n",
"outputId": "12f20f62-9021-4523-cbc6-ec172a6322d7"
},
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Sigmoid outputs: tensor([0.88, 0.27, 0.62])\n"
]
}
],
"source": [
"import torch\n",
"# import torch.nn.functional as F\n",
"\n",
"# 원시 출력값 (logits)\n",
"logits = torch.tensor([2.0, -1.0, 0.5])\n",
"\n",
"# Sigmoid 적용\n",
"sigmoid_outputs = torch.sigmoid(logits)\n",
"\n",
"# Print Output\n",
"print('Sigmoid outputs:', sigmoid_outputs)"
]
},
{
"cell_type": "code",
"source": [
"import torch\n",
"import torch.nn.functional as F\n",
"\n",
"\n",
"# 원시 출력값 (logits)\n",
"logits = torch.tensor([2.0, 1.0, 0.1])\n",
"\n",
"# Softmax 적용\n",
"softmax_outputs = F.softmax(logits, dim=0)\n",
"print(f'{softmax_outputs=}')"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "FA4lK-wyRzRo",
"outputId": "1a56c26d-f97d-4542-a557-9a468ecf5e89"
},
"execution_count": 31,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"softmax_outputs=tensor([0.66, 0.24, 0.10])\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import torch\n",
"# Softmax 모듈 초기화 (dim=-1은 텐서의 마지막 차원을 기준으로 softmax를 적용)\n",
"softmax = torch.nn.Softmax(dim=-1)\n",
"\n",
"\n",
"# 원시 출력값 (logits)\n",
"logits = torch.tensor([2.0, 1.0, 0.1])\n",
"\n",
"# Softmax 적용.\n",
"softmax_outputs = softmax(logits)\n",
"print(f'{softmax_outputs=}')"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "bqb0gA26UfbE",
"outputId": "1e048dce-6b56-4a3a-eb09-51033e8a73a4"
},
"execution_count": 32,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"softmax_outputs=tensor([0.66, 0.24, 0.10])\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import torch.nn as nn\n",
"\n",
"# 예측값과 실제값\n",
"outputs = torch.tensor([0.7, 0.2, 0.9], requires_grad=True).float()\n",
"targets = torch.tensor([1.0, 0.0, 1.0]).float()\n",
"\n",
"# BCELoss 선언\n",
"criterion = nn.BCELoss()\n",
"\n",
"# 시그모이드 활성화 적용\n",
"outputs = torch.sigmoid(outputs)\n",
"\n",
"# 손실 계산\n",
"loss = criterion(outputs, targets)\n",
"print('BCELoss:', loss.item())"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "5kJYII7gU9Lt",
"outputId": "9e71592d-5c7a-47ba-91d9-0001a5ab388e"
},
"execution_count": 33,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"BCELoss: 0.514159619808197\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import torch\n",
"import torch.nn as nn\n",
"\n",
"# 원시 출력값 (시그모이드 활성화 적용 전 raw score)\n",
"raw_outputs = torch.tensor([0.5, -1.0, 2.0], requires_grad=True)\n",
"targets = torch.tensor([1.0, 0.0, 1.0])\n",
"\n",
"# BCEWithLogitsLoss 선언\n",
"criterion = nn.BCEWithLogitsLoss()\n",
"\n",
"# 손실 계산 (시그모이드 내부 적용)\n",
"loss = criterion(raw_outputs, targets)\n",
"print(f'BCEWithLogitsLoss: {loss.item():.4f}')"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "GIrTV8P_ZfAN",
"outputId": "bb8c69b8-5b6d-4d0e-c951-c03babab1c5d"
},
"execution_count": 34,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"BCEWithLogitsLoss: 0.3048\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"# 원시 출력값 (softmax 활성화 적용 전)\n",
"raw_outputs = torch.tensor(\n",
"\t[[1.0, 2.0, 3.0],\n",
" [1.0, 2.0, 0.0],\n",
" [0.0, 2.0, 1.0]],\n",
" requires_grad=True).float()\n",
"targets = torch.tensor([2, 0, 1]).long() # 각 샘플의 클래스 인덱스\n",
"print(targets.dtype)\n",
"# CrossEntropyLoss 선언\n",
"criterion = nn.CrossEntropyLoss()\n",
"\n",
"# 손실 계산 (softmax 내부 적용)\n",
"loss = criterion(raw_outputs, targets)\n",
"print(f'CrossEntropyLoss: {loss.item():.2f}')"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "-UYH_zVka0t3",
"outputId": "1dbb2a0a-0326-41b6-b375-7963eb59c263"
},
"execution_count": 35,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"torch.int64\n",
"CrossEntropyLoss: 0.74\n"
]
}
]
},
{
"cell_type": "code",
"source": [
"import torch\n",
"import torch.nn\n",
"\n",
"# 모델 출력\n",
"logits = torch.tensor(\n",
"\t[[0.1, 0.2, 0.7], # max idx 2\n",
" \t [0.8, 0.1, 0.1], # max idx 0\n",
" \t [0.3, 0.5, 0.2]], # max idx 1\n",
" \trequires_grad=True).float()\n",
"\n",
"# 타겟 클래스 인덱스\n",
"targets = torch.tensor([2, 0, 1]).long()\n",
"\n",
"# 로그 소프트맥스 적용\n",
"log_softmax = F.log_softmax(logits, dim=1)\n",
"\n",
"# NLLLoss 선언\n",
"criterion = nn.NLLLoss()\n",
"loss = criterion(log_softmax, targets)\n",
"print(f'NLLLoss with LogSoftmax: {loss.item():.2f}')"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "EJZExLiBbsyC",
"outputId": "c4e878a9-e60e-4aa5-c48b-64959243b80c"
},
"execution_count": 36,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"NLLLoss with LogSoftmax: 0.80\n"
]
}
]
}
]
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment