删除恒源云的使用代码;修改北京云的训练脚本; main中修改了部分参数并在设置gpu的情况下自动调用gpu; 注明train model中训练步的返回值

free_servers/bscc-run.sh

@@ -1,14 +1,14 @@
 #!/bin/bash
 
-module load anaconda/2020.11
-module load cuda/10.2
-module load cudnn/8.1.1.33_CUDA10.2
+#cd run
+#cd machineLearningTemplate
 
-#conda create --name py37 python=3.7
-source activate py37
+module load anaconda/2020.11
+module load cuda/11.1
+module load cudnn/8.1.1.33_CUDA11.1
 
-cd run
-cd machineLearningScaffold
+#conda create --name py39 python=3.9
+source activate py39
 
 #pip install -r requirements.txt
 

free_servers/hy-scaffold.ipynb

@@ -1,92 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "4c4d04db-6fe0-4511-abb3-c746ba869863",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "!apt-get update\n",
-    "!apt-get install p7zip-full -y"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "7c027a62-8116-4c55-9a90-95c4d90dd087",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "!7z x dataset.zip -o/hy-tmp\n",
-    "!7z x machineLearningScaffold.zip -o/hy-tmp\n",
-    "!7z x vit_pre-train_checkpoint.zip -o/hy-tmp\n",
-    "!mv -f /hy-tmp/pre-train_checkpoint/vit_checkpoint/imagenet21k/R50+ViT-B_16.npz /hy-tmp/pre-train_checkpoint"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 8,
-   "id": "4c6dc6a1-0208-4efb-beaf-b1264a94ba71",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "\u001b[33mWARNING: You are using pip version 21.0.1; however, version 21.2.4 is available.\n",
-      "You should consider upgrading via the '/usr/bin/python3.8 -m pip install --upgrade pip' command.\u001b[0m\n"
-     ]
-    }
-   ],
-   "source": [
-    "!pip install -r /hy-tmp/requirements.txt > /dev/null"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "2c074f52-d27e-4605-8b4f-45b11a2f1eb1",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import main\n",
-    "%load_ext tensorboard"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "a5a8d64e-d90e-467f-a3b4-db8b31874044",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "main.main('fit', gpus=1, dataset_path='./dataset/MFNet(RGB-T)', num_workers=2, max_epochs=60, batch_size=16, precision=16, seed=1234, # tpu_cores=8,\n",
-    "     checkpoint_every_n_val=1, save_name='version_0', #checkpoint_path='/version_15/checkpoints/epoch=59-step=4739.ckpt',\n",
-    "     path_final_save='./drive/MyDrive', save_top_k=1\n",
-    "     )"
-   ]
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "Python 3",
-   "language": "python",
-   "name": "python3"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3",
-   "version": "3.8.10"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 5
-}

main.py

@@ -1,3 +1,5 @@
+import torch
+
 from save_checkpoint import SaveCheckpoint
 from data_module import DataModule
 from pytorch_lightning import loggers as pl_loggers
@@ -27,6 +29,11 @@ def main(stage,
     框架的入口函数. 包含设置超参数, 划分数据集, 选择训练或测试等流程
     该函数的参数为训练过程中需要经常改动的参数
 
+    经常改动的      参数    作为main的输入参数
+    不常改动的   非通用参数    存放在config
+    不常改动的    通用参数     直接进行声明
+    * 通用参数指的是所有网络中共有的参数, 如time_sum等
+
     :param stage: 表示处于训练阶段还是测试阶段, fit表示训练, test表示测试
     :param max_epochs:
     :param batch_size:
@@ -44,14 +51,11 @@ def main(stage,
                            非重载训练的情况下, 可以通过调整该值控制训练的次数;
     :param k_fold:
     """
-    # 经常改动的      参数    作为main的输入参数
-    # 不常改动的   非通用参数    存放在config
-    # 不常改动的    通用参数     直接进行声明
-    # 通用参数指的是所有网络中共有的参数, 如time_sum等
-
     # 处理输入数据
     precision = 32 if (gpus is None and tpu_cores is None) else precision
-    # 获得通用参数
+    # 自动处理:param gpus
+    gpus = 1 if torch.cuda.is_available() and gpus is None and tpu_cores is None else None
+    # 定义不常改动的通用参数
     # TODO 获得最优的batch size
     num_workers = cpu_count()
     # 获得非通用参数
@@ -100,8 +104,7 @@ def main(stage,
 
 
 if __name__ == "__main__":
-    main('fit', max_epochs=2, batch_size=32, precision=16, seed=1234, dataset_path='./dataset', k_fold=10,
+    main('fit', max_epochs=200, batch_size=128, precision=16, seed=1234, dataset_path='./dataset', k_fold=10,
          kth_fold_start=9,
-         # gpus=1,
          # version_nth=8,
          )

train_model.py

@@ -17,6 +17,7 @@ class TrainModule(pl.LightningModule):
         self.net = resnet56()
         self.loss = nn.CrossEntropyLoss()
 
+    # 返回值必须包含loss, loss可以作为dict中的key, 或者直接返回loss
     def training_step(self, batch, batch_idx):
         _, input, label = batch
         label = label.flatten()