[ML] AutoML 功能比較
===
###### tags: `ML`, `Azure`, `Google`

<br>

<table>
<tr>
<th colspan="2">功能</th>
<th>Azure</th>
<th>Google</th>
<th>AWS</th>
<th>說明</th>
</tr>
<tr>
<td colspan="6"><b>資料集</b></td>
</tr>
<tr>
<td rowspan="4">資料來源設定</td>
<td><li style="margin-left:0.5em">本機</li></td>
<td><input type="checkbox" checked /></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td></td>
</tr>
<tr>
<td><li style="margin-left:0.5em">儲存體</li></td>
<td><input type="checkbox" checked /></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td></td>
</tr>
<tr>
<td><li style="margin-left:0.5em">Web</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>
<tr>
<td style="white-space: nowrap"><li style="margin-left:0.5em">公開資料集</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>
<tr>
<td rowspan="2">資料集型式</td>
<td><li style="margin-left:0.5em">表格型</li></td>
<td><input type="checkbox" checked /></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td></td>
</tr>
<tr>
<td><li style="margin-left:0.5em">非表格型</li></td>
<td><input type="checkbox" checked /> ①</td>
<td><input type="checkbox" checked />
<br>- 文字<br>- 圖片<br>- 影片
</td>
<td></td>
<td>① 檔案資料集(非表格型)，需透過 "設計工具 (pipeline)" 進行資料格式轉換？</td>
</tr>
<tr>
<td rowspan="1">資料內容檢視<br>(表格型)</td>
<td><li style="margin-left:0.5em">預覽部份資料</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td>預覽<br>- 前50筆<br>- 前1000筆</td>
</tr>
<tr>
<td rowspan="2">資料內容分析<br>(表格型)</td>
<td><li style="margin-left:0.5em">各欄位分佈概況</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>
<tr>
<td><li style="margin-left:0.5em">各欄位基本統計</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td>- 時刻數(平均值, 標準差)<br>- 分位數<br>- 儲存格計數</td>
</tr>
<tr>
<td colspan="6"></td>
</tr>

<tr>
<td colspan="6"><b>自動化 ML</b></td>
</tr>

<tr>
<td rowspan="6">支援任務</td>
<td><li style="margin-left:0.5em">表格型：分類</li></td>
<td><input type="checkbox" checked /></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td><a href="https://docs.microsoft.com/en-us/azure/machine-learning/how-to-configure-auto-train#supported-models" target="_blank">Azure 演算法</a>
</td>
</tr>

<tr>
<td><li style="margin-left:0.5em">表格型：迴歸</li></td>
<td><input type="checkbox" checked /></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td><a href="https://docs.microsoft.com/en-us/azure/machine-learning/how-to-configure-auto-train#supported-models" target="_blank">Azure 演算法</a>
</td>
</tr>

<tr>
<td style="white-space: nowrap"><li style="margin-left:0.5em">表格型：時間序列預測</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td><a href="https://docs.microsoft.com/en-us/azure/machine-learning/how-to-configure-auto-train#supported-models" target="_blank">Azure 演算法</a>
</td>
</tr>

<tr>
<td><li style="margin-left:0.5em">圖片</li></td>
<td></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td><a href="https://i.imgur.com/LDuVTAY.png" target="_blank">圖片</a><br>- 單一標籤<br>- 多標籤<br>- 物件偵測<br>- 實體切割</td>
</tr>

<tr>
<td><li style="margin-left:0.5em">文字</li></td>
<td></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td><a href="https://i.imgur.com/ap9fxQW.png" target="_blank">文字</a><br>- 單一標籤<br>- 多標籤<br>- 實體擷取<br>- 情緒分析</td>
</tr>

<tr>
<td><li style="margin-left:0.5em">影片</li></td>
<td></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td><a href="https://i.imgur.com/7tRp4Es.png" target="_blank">影片</a><br>- <a href="https://www.gstatic.com/pantheon/images/automl/objective_video_action_recognition.gif" target="_blank">動作搜尋</a><br>- <a href="https://www.gstatic.com/pantheon/images/automl/objective_video_classification.gif" target="_blank">影片分類</a><br>- <a href="https://www.gstatic.com/pantheon/images/automl/objective_video_object_tracking.gif" target="_blank">物件追蹤</a></td>
</tr>

<tr>
<td>自訂訓練</td>
<td><li style="margin-left:0.5em">自訂 ML 框架
<ul style="margin-left:0.5em">
<li>PyTorch
<li>scikit-learn
<li>TensorFlow
<li>XGBoost
<li>自訂容器
</ul>
</li>
<li>超參數調整 (選用)</li>
</td>
<td>❎</td>
<td><input type="checkbox" checked /></td>
<td></td>
<td>Google:
<br><a href="https://cloud.google.com/ai-platform-unified/docs/training/containers-overview" target="_blank">使用指南</a></td>
</tr>

<tr>
<td rowspan="5">組態設定<br>(表格型)</td>
<td><li style="margin-left:0.5em">選擇模型衡量指標</li></td>
<td><input type="checkbox" checked />
<small><small><ul style="margin-left:0.5em">
<li><b>Accuracy</b>
<li>AUC weighted
<li>Norm macro recall
<li>Average precision score weighted
<li>Precision score weighted
</ul></small></small>
</td>
<td><input type="checkbox" checked />
<small><small><ul style="margin-left:0.5em">
<li>AUC PRC
<li>AUC ROC
<li><b>Log loss</b>
<li>Precision
<li>Recall
</ul></small></small>
</td>
<td></td>
<td></td>
</tr>

<tr>
<td><li style="margin-left:0.5em">排除演算法</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>

<tr>
<td><li style="margin-left:0.5em">設定訓練結束條件<br>(Early stopping)</li></td>
<td><input type="checkbox" checked />
<small><small><ul style="margin-left:0.5em">
<li>執行總時間
<li>滿足指標門檻
</ul></small></small>
</td>
<td><input type="checkbox" checked />
<small><small><ul style="margin-left:0.5em">
<li>沒有更多<br>的改善
</ul></small></small>
</td>
<td></td>
<td></td>
</tr>

<tr>
<td><li style="margin-left:0.5em">資料集的驗證方式：
<small><small>
<ul style="margin-left:0.5em">
<li>K 折交叉驗證(k=5,固定)
<li>蒙地卡羅交叉驗證(k=5,隨機)
<li>訓練/驗證分割
<li>訓練/驗證/測試分割
</ul>
</small></small>
</li></td>
<td><br>
<small><small>
<ul style="margin-left:0.5em">
<li><input type="checkbox" checked />
<li><input type="checkbox" checked />
<li><input type="checkbox" checked />
<li>❎<br><br>
</ul>
</small></small>
</td>
<td><br>
<small><small>
<ul style="margin-left:0.5em">
<li>❎
<li>❎
<li>❎
<li><input type="checkbox" checked /><br><br>
</ul>
</small></small>
</td>
<td></td>
<td></td>
</tr>

<tr>
<td><li style="margin-left:0.5em">資料缺失與插捕方式：
<ul style="margin-left:0.5em">
<li>自動
<li>平均數
<li>最常見
<li>中位數
<li>填入常數
</ul>
</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>

<tr>
<td rowspan="2">報表呈現<br>(表格型)</td>
<td><li style="margin-left:0.5em">模型排行榜</li></td>
<td><input type="checkbox" checked />
<br>(多個模型)
</td>
<td>❎<br>(單一模型)</td>
<td></td>
<td></td>
</tr>
<tr>
<td><li style="margin-left:0.5em">資料集分析報告
<ul style="margin-left:0.5em">
<li>目標類別平衡偵測
<li>高基數特徵偵測
<li>遺漏特徵值插補
</ul>
</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>
<tr>
<td colspan="6"></td>
</tr>

<tr>
<td colspan="6"><b>模型評估</b></td>
</tr>
<tr>
<td>檢視模型各指標</td>
<td><li style="margin-left:0.5em">(你想到的各種指標)</li></td>
<td>
<input type="checkbox" checked />
<br>all
</td>
<td style="white-space: nowrap">
<input type="checkbox" checked />
<br>- PR AUC
<br>- ROC AUC
<br>- Log loss
<br>- F1
<br>- Precision
<br>- Recall
</td>
<td></td>
<td></td>
</tr>

<tr>
<td rowspan="3"><a href="https://docs.microsoft.com/zh-tw/azure/machine-learning/how-to-machine-learning-interpretability-aml" target="_blank">可解釋性分析</a></td>
<td><li style="margin-left:0.5em">特徵重要性</li></td>
<td><input type="checkbox" checked /></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td></td>
</tr>

<tr>
<td><li style="margin-left:0.5em">任兩欄位資料分佈</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>

<tr>
<td><li style="margin-left:0.5em">任兩欄位重要性關聯</li></td>
<td><input type="checkbox" checked /></td>
<td>❎</td>
<td></td>
<td></td>
</tr>

<tr>
<td colspan="6"></td>
</tr>

<tr>
<td colspan="6"><b>模型佈署與端點連接</b></td>
</tr>

<tr>
<td rowspan="2">REST API</td>
<td>驗證方式</td>
<td><input type="checkbox" checked />
<br>- 不使用
<br>- 金鑰式
<br>- 權杖式
</td>
<td><input type="checkbox" checked />
<br>(Google<br>代管金鑰)
</td>
<td></td>
<td></td>
</tr>

<tr>
<td>API 使用範例</td>
<td><input type="checkbox" checked />
<br>- C#
<br><span style="white-space: nowrap;">- Python</span>
<br>- R
</td>
<td><input type="checkbox" checked />
<br>- curl
<br><span style="white-space: nowrap;">- Python</span>
<br>
<br>
</td>
<td></td>
<td>Google: 資料內容需自行填寫</td>
</tr>

<tr>
<td rowspan="2">模型線上測試</td>
<td><li style="margin-left:0.5em">單筆測試 (透過端點)</li></td>
<td><input type="checkbox" checked /></td>
<td><input type="checkbox" checked /></td>
<td></td>
<td></td>
</tr>

<tr>
<td><li style="margin-left:0.5em">批次測試 (非透過端點)
</li>
</td>
<td>❎</td>
<td><input type="checkbox" checked />
<br>(上傳檔案)
</td>
<td></td>
<td></td>
</tr>

</table>

<br>

<hr>

<br>

## 參考資料
- [Amazon SageMaker vs Azure Machine Learning Studio](https://www.trustradius.com/compare-products/amazon-sagemaker-vs-azure-machine-learning-studio)
- Pros v.s. Cons

### Azure
- [Make big data and AI easy with Microsoft Azure Databricks](https://www.softwebsolutions.com/resources/integration-of-azure-databricks.html)

### Google
- [[Google] AutoML Tables (BETA 版)](https://cloud.google.com/automl-tables)
- [[Day 26] Google AutoML Table - 1](https://ithelp.ithome.com.tw/articles/10224282?sc=iThelpR)