Dev

DESIGN_DOC_UI_UX.md

@@ -0,0 +1,33 @@
+### 2.1. 模型管理 (Model Management)
+
+*   **目的**: 创建、训练、管理和评估用于预测的机器学习模型。产出可供策略使用的"已训练模型"。
+*   **主要页面**:
+    *   **`models.html` (模型列表页)**:
+        *   **功能**: 展示所有已创建的模型及其状态和关键信息。提供新建模型入口。
+        *   **表格列**:
+            *   名称
+            *   类型 (LightGBM, LSTM, ...)
+            *   **训练周期** (例如: "2018-01-01 - 2022-12-31")
+            *   **训练股票池** (例如: "CSI 300")
+            *   **特征集** (例如: "Alpha158", 使用 `truncate` 显示，`title` 显示完整内容)
+            *   **标签** (例如: "Ref($close,-2)/Ref($close,-1)-1", 使用 `truncate` 显示，`title` 显示完整内容)
+            *   状态 (草稿, 训练中..., 已训练, 失败)
+            *   创建时间
+            *   操作 (详情/编辑, 训练/重新训练, 删除, 取消训练等，根据状态显示不同按钮)
+    *   **`create_model.html` (新建模型页)**:
+        *   **功能**: 配置并启动一个新的模型训练任务。
+        *   **卡片/区域**:
+            *   **基本信息**: 模型名称 (必填), 模型描述。
+            *   **训练配置 (Training Configuration - 新增)**:
+                *   训练开始日期 (必填, `date`)
+                *   训练结束日期 (必填, `date`)
+                *   训练股票池 (必填, `select`, 包含 CSI 300, CSI 500, 全市场A股等)
+                *   所用特征集 (必填, `text`, placeholder: "例如: Alpha158, VWAP")
+                *   标签/预测目标 (必填, `text`, placeholder: "例如: Ref($close,-2)/Ref($close,-1)-1")
+            *   **模型配置 (Model Configuration)**:
+                *   模型类型 (必填, `select`, LightGBM, LSTM, ...)
+                *   模型参数 (根据类型动态显示常用参数输入，或提供高级 YAML/JSON 配置入口)
+            *   **数据配置 (Data Configuration - 已精简)**: *此部分原始目的是映射 Qlib DataHandler 配置。为避免与"训练配置"重复，已移除标的范围(Universe)、特征、标签、训练期、验证期等字段。当前仅保留"数据源"和"数据频率"字段，用于定义基础数据访问信息。其最终必要性待后续审视。*
+        *   **操作按钮**: 取消, 保存草稿, **保存并开始训练**。
+    *   **`edit_model.html` (编辑/详情页)**:
+        *   **功能**: 查看模型详细配置、训练状态、训练日志（未来）、评估结果（未来），并可编辑配置（对于草稿状态）或触发重新训练。 

examples/benchmarks/LSTM/workflow_config_lstm_Alpha158.yaml

@@ -17,18 +17,14 @@ data_handler_config: &data_handler_config
                             "ROC60", "RESI10", "VSTD5", "RSQR60", "CORR60", "WVMA60", "STD5", 
                             "RSQR20", "CORD60", "CORD10", "CORR20", "KLOW"
                         ]
-        - class: RobustZScoreNorm
+        - class: ZScoreNorm
           kwargs:
               fields_group: feature
-              clip_outlier: true
         - class: Fillna
           kwargs:
               fields_group: feature
     learn_processors:
         - class: DropnaLabel
-        - class: CSRankNorm
-          kwargs:
-              fields_group: label
     label: ["Ref($close, -2) / Ref($close, -1) - 1"]
 
 port_analysis_config: &port_analysis_config
@@ -59,13 +55,13 @@ task:
             hidden_size: 64
             num_layers: 2
             dropout: 0.0
-            n_epochs: 200
-            lr: 1e-3
+            n_epochs: 20
+            lr: 1e-5
             early_stop: 10
             batch_size: 800
             metric: loss
             loss: mse
-            n_jobs: 20
+            n_jobs: 8
             GPU: 0
     dataset:
         class: TSDatasetH

my_work/cursor_rules.md

@@ -0,0 +1,23 @@
+# Cursor AI Assistant Rules & Context
+
+## Project Goal
+Learn the Qlib framework for quantitative trading by studying its source code.
+
+## Qlib Source Location
+The main Qlib source code is located at the workspace root: `/Users/chengzheng/workspace/trade/qlib`
+
+## User Workspace
+My personal learning scripts and experiments related to Qlib will be created and stored within the `/Users/chengzheng/workspace/trade/qlib/my_work/` directory. Please place any new scripts or generated files here unless otherwise specified.
+
+## Learning Objectives
+- Master the application of all algorithms within Qlib for quantitative trading, including strategy design and backtesting.
+- Select and implement specific AI algorithms (learned through Qlib) to build custom trading strategies.
+- Gain knowledge in Deep Learning (DL) and Reinforcement Learning (RL) concepts, particularly their application in quantitative finance, potentially using Qlib examples.
+- Understand the overall architecture and design principles of the Qlib framework, referencing official documentation (e.g., [Qlib Quick Start](https://www.wuzao.com/document/qlib/introduction/quick.html)) and source code, to enable effective modular usage and customization.
+
+## Interaction Style
+- Assume I have access to the Qlib source code in the specified location.
+- When discussing code, refer to file paths relative to the Qlib root (`/Users/chengzheng/workspace/trade/qlib`).
+- When creating new code or examples for me, place them in the `my_work/` directory.
+- Follow the RIPER-5 protocol for structured interaction.
+- Prioritize Chinese for general conversation, while adhering to the protocol's format requirements for specific elements (e.g., mode declarations, checklists).