Skip to content

Commit

Permalink
Merge pull request #4 from XinshaoAmosWang/DCL-11-add-jupyter-noteboo…
Browse files Browse the repository at this point in the history 
…k-for-training-shufflenet-resnet-18-on-cifar-100

DCL-11: add a jupyter notebook demo for training shufflenetv2 on cifar-100 with a symmetric noise rate of 40%
  • Loading branch information
@XinshaoAmosWang
XinshaoAmosWang authored Oct 29, 2022
2 parents df5bf58 + 85c5af3 commit ea71596
Show file tree
Hide file tree
Showing 2 changed files with 314 additions and 2 deletions.
9 changes: 7 additions & 2 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -57,6 +57,9 @@ Here are related papers on the fitting and generalization of deep learning:
* Easy to run experiments and sink results
* Easy to put sinked results into your technical reports and academic papers.

## Demos
* [Training the shufflenetv2 on cifar-100 with a symmetric noise rate of 40%](./demos_jupyter_notebooks/convnets_cifar100/trainer_cifar100_covnets_proselflc.ipynb)

## Install

<details><summary>See Install Guidelines</summary>
Expand Down Expand Up @@ -131,6 +134,8 @@ The results are well sinked and organised, e.g.,
#### [Link to CVPR 2021 Slide, Poster](./Poster_Slide/CVPR-2021)
#### [Link to reviewers' comments](./Reviews)

## Notes
## Notes and disclaimer
* For any specific research discussion or potential future collaboration, please feel free to contact me. <br />
* This work is a personal research project and in progress outside of company work.
* This work is a personal research project and in progress using personal time.

## [LICENSE](LICENSE)
307 changes: 307 additions & 0 deletions demos_jupyter_notebooks/convnets_cifar100/trainer_cifar100_covnets_proselflc.ipynb
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,307 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"\n",
"## Import dependencies"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import time\n",
"\n",
"import pandas\n",
"import torch\n",
"\n",
"from proselflc.trainer.trainer_cnn_vision_derivedgrad import Trainer\n",
"\n",
"import pprint\n",
"print = pprint.PrettyPrinter(indent=4).pprint"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Add params configurations"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{ 'batch_accumu_steps': 10,\n",
" 'batch_size': 128,\n",
" 'classes_per_batch': 64,\n",
" 'counter': 'iteration',\n",
" 'data_name': 'cifar100',\n",
" 'device': 'gpu',\n",
" 'eval_interval': 500,\n",
" 'exp_base': 16,\n",
" 'gamma': 0.1,\n",
" 'logit_soften_T': 0.5,\n",
" 'loss_mode': 'cross entropy',\n",
" 'loss_name': 'proselflc',\n",
" 'lr': 0.2,\n",
" 'lr_scheduler': 'WarmupMultiStepSchedule',\n",
" 'milestones': [20000, 30000],\n",
" 'momentum': 0.9,\n",
" 'network_name': 'shufflenetv2',\n",
" 'num_classes': 100,\n",
" 'num_workers': 8,\n",
" 'sampler': 'BalancedBatchSampler',\n",
" 'symmetric_noise_rate': 0.4,\n",
" 'total_epochs': 100,\n",
" 'transit_time_ratio': 0.5,\n",
" 'trust_mode': 'global*(1-H(p)/H(u))',\n",
" 'warmup_epochs': 0,\n",
" 'weight_decay': 0.001}\n"
]
}
],
"source": [
"params = {}\n",
"params.update(\n",
" {\n",
" \"data_name\": \"cifar100\",\n",
" \"num_classes\": 100, # 1000\n",
" \"device\": \"gpu\",\n",
" #\n",
" \"num_workers\": 8,\n",
" #\n",
" \"counter\": \"iteration\",\n",
" \"classes_per_batch\": 8,\n",
" #\n",
" # To get deterministic results, please uncomment the line below\n",
" # to set seed parameter.\n",
" # \"seed\": 123,\n",
" }\n",
")\n",
"\n",
"# data\n",
"params[\"symmetric_noise_rate\"] = 0.4\n",
"\n",
"# network\n",
"params[\"network_name\"] = \"shufflenetv2\"\n",
"\n",
"# for the demo purpose, I set the total epochs to be small.\n",
"params[\"total_epochs\"] = 100\n",
"params[\"eval_interval\"] = 500 # iterations\n",
"\n",
"# batch\n",
"params[\"batch_size\"] = 128\n",
"params[\"sampler\"] = \"BalancedBatchSampler\"\n",
"params[\"classes_per_batch\"] = 64\n",
"\n",
"# learning rate\n",
"params[\"lr\"] = 0.2\n",
"params[\"weight_decay\"] = 1e-3\n",
"params[\"lr_scheduler\"] = \"WarmupMultiStepSchedule\"\n",
"params[\"warmup_epochs\"] = 0\n",
"params[\"milestones\"] = [20000, 30000]\n",
"\n",
"# optimisation\n",
"params[\"momentum\"] = 0.9\n",
"params[\"batch_accumu_steps\"] = 10\n",
"params[\"gamma\"] = 0.1\n",
"\n",
"# loss settings\n",
"params[\"loss_mode\"] = \"cross entropy\"\n",
"params[\"trust_mode\"] = \"global*(1-H(p)/H(u))\"\n",
"params[\"loss_name\"] = \"proselflc\"\n",
"params[\"transit_time_ratio\"] = 0.50\n",
"params[\"exp_base\"] = 16\n",
"params[\"logit_soften_T\"] = 0.5\n",
"\n",
"print(params)"
]
},
{
"cell_type": "markdown",
"source": [
"## Create the folder to store intermediate and final results\n",
"* First, run `sudo mkdir /home/proselflc_experiments/ && sudo chmod -R 777 /home/proselflc_experiments/` so that so you have the write permission.\n",
"* Or set `WORK_DIR = \"/home/your_username/proselflc_experiments/\"`"
],
"metadata": {
"collapsed": false
}
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [],
"source": [
"WORK_DIR = \"/home/proselflc_experiments/\"\n",
"#\n",
"# use the time as a unique experiment identifier\n",
"dt_string = time.strftime(\"%Y%m%d-%H%M%S\")\n",
"summary_writer_dir = (\n",
" params[\"loss_name\"]\n",
" + \"_\"\n",
" + dt_string\n",
")\n",
"params[\"summary_writer_dir\"] = (\n",
" WORK_DIR\n",
" + \"/\"\n",
" + params[\"data_name\"]\n",
" + \"_symmetric_noise_rate_\"\n",
" + str(params[\"symmetric_noise_rate\"])\n",
" + \"/\"\n",
" + params[\"network_name\"]\n",
" + \"/\"\n",
" + summary_writer_dir\n",
")\n",
"if not os.path.exists(params[\"summary_writer_dir\"]):\n",
" os.makedirs(params[\"summary_writer_dir\"])"
]
},
{
"cell_type": "markdown",
"source": [
"## Init the trainer, and store the params configurations\n",
"For each experiment, we have one unique result folder to store the params configurations and learning curves.\n",
"Therefore, you can revisit any specific experiment whenever you need without losing any details."
],
"metadata": {
"collapsed": false
}
},
{
"cell_type": "code",
"execution_count": 9,
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Files already downloaded and verified\n",
"Files already downloaded and verified\n",
"Files already downloaded and verified\n"
]
}
],
"source": [
"trainer = Trainer(params=params)\n",
"# params[\"milestones\"] was a list of integers, we convert it to a string before sinking.\n",
"params[\"milestones\"] = str(params[\"milestones\"])\n",
"dataframe = pandas.DataFrame(params, index=[0])\n",
"dataframe.to_csv(\n",
" params[\"summary_writer_dir\"] + \"/params.csv\",\n",
" encoding=\"utf-8\",\n",
" index=False,\n",
" sep=\"\\t\",\n",
" mode=\"w\", #\n",
")"
],
"metadata": {
"collapsed": false
}
},
{
"cell_type": "markdown",
"source": [
"## Run the trainer and save the final model"
],
"metadata": {
"collapsed": false
}
},
{
"cell_type": "code",
"execution_count": 10,
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Evaluating Network.....\n",
"Iteration= (500,2)/(39100, 100), lr=0.2000, batch_mean_epsilon=0.0001, valid batch size=128.0000, batch_mean_gtrust=0.0004, batch_mean_etrust=0.1395\n",
"clean_test: Loss= 4.2554, Accuracy= 0.0538, Entropy= 0.9492, Max_p= 0.0478,\n",
"noisy_subset: Loss= 4.7921, Accuracy= 0.0101, Entropy= 0.9555, Max_p= 0.0434,\n",
"clean_subset: Loss= 4.2743, Accuracy= 0.0535, Entropy= 0.9552, Max_p= 0.0437,\n",
"cleaned_noisy_subset: Loss= 4.2758, Accuracy= 0.0546, Entropy= 0.9554, Max_p= 0.0435,\n",
"Evaluating Network.....\n",
"Iteration= (1000,3)/(39100, 100), lr=0.2000, batch_mean_epsilon=0.0001, valid batch size=128.0000, batch_mean_gtrust=0.0005, batch_mean_etrust=0.2463\n",
"clean_test: Loss= 3.9042, Accuracy= 0.1096, Entropy= 0.9200, Max_p= 0.0681,\n",
"noisy_subset: Loss= 4.8793, Accuracy= 0.0092, Entropy= 0.9332, Max_p= 0.0584,\n"
]
},
{
"ename": "KeyboardInterrupt",
"evalue": "",
"output_type": "error",
"traceback": [
"\u001B[0;31m---------------------------------------------------------------------------\u001B[0m",
"\u001B[0;31mKeyboardInterrupt\u001B[0m Traceback (most recent call last)",
"Input \u001B[0;32mIn [10]\u001B[0m, in \u001B[0;36m<cell line: 1>\u001B[0;34m()\u001B[0m\n\u001B[0;32m----> 1\u001B[0m \u001B[43mtrainer\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mtrain\u001B[49m\u001B[43m(\u001B[49m\u001B[43m)\u001B[49m\n\u001B[1;32m 2\u001B[0m torch\u001B[38;5;241m.\u001B[39msave(\n\u001B[1;32m 3\u001B[0m trainer\u001B[38;5;241m.\u001B[39mnetwork,\n\u001B[1;32m 4\u001B[0m params[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124msummary_writer_dir\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m+\u001B[39m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124m/model.pt\u001B[39m\u001B[38;5;124m\"\u001B[39m,\n\u001B[1;32m 5\u001B[0m )\n\u001B[1;32m 6\u001B[0m \u001B[38;5;28mprint\u001B[39m(\n\u001B[1;32m 7\u001B[0m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mThe experiment is finished with details sinked in \u001B[39m\u001B[38;5;132;01m{}\u001B[39;00m\u001B[38;5;124m\"\u001B[39m\u001B[38;5;241m.\u001B[39mformat(\n\u001B[1;32m 8\u001B[0m params[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124msummary_writer_dir\u001B[39m\u001B[38;5;124m\"\u001B[39m]\n\u001B[1;32m 9\u001B[0m )\n\u001B[1;32m 10\u001B[0m )\n",
"File \u001B[0;32m~/Dropbox/GeneralCareerDevelop/PersonalBrand/GithubRepositories/ProSelfLC_tpami2release/src/proselflc/trainer/trainer_cnn_vision_derivedgrad.py:258\u001B[0m, in \u001B[0;36mTrainer.train\u001B[0;34m(self)\u001B[0m\n\u001B[1;32m 254\u001B[0m \u001B[38;5;28;01mdef\u001B[39;00m \u001B[38;5;21mtrain\u001B[39m(\u001B[38;5;28mself\u001B[39m) \u001B[38;5;241m-\u001B[39m\u001B[38;5;241m>\u001B[39m \u001B[38;5;28;01mNone\u001B[39;00m:\n\u001B[1;32m 255\u001B[0m \u001B[38;5;66;03m# #############################\u001B[39;00m\n\u001B[1;32m 256\u001B[0m \u001B[38;5;28;01mfor\u001B[39;00m epoch \u001B[38;5;129;01min\u001B[39;00m \u001B[38;5;28mrange\u001B[39m(\u001B[38;5;241m1\u001B[39m, \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mtotal_epochs \u001B[38;5;241m+\u001B[39m \u001B[38;5;241m1\u001B[39m):\n\u001B[1;32m 257\u001B[0m \u001B[38;5;66;03m# train one epoch\u001B[39;00m\n\u001B[0;32m--> 258\u001B[0m \u001B[38;5;28;43mself\u001B[39;49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mtrain_one_epoch\u001B[49m\u001B[43m(\u001B[49m\n\u001B[1;32m 259\u001B[0m \u001B[43m \u001B[49m\u001B[43mepoch\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mepoch\u001B[49m\u001B[43m,\u001B[49m\n\u001B[1;32m 260\u001B[0m \u001B[43m \u001B[49m\u001B[43mdataloader\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[38;5;28;43mself\u001B[39;49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mtraindataloader\u001B[49m\u001B[43m,\u001B[49m\n\u001B[1;32m 261\u001B[0m \u001B[43m \u001B[49m\u001B[43m)\u001B[49m\n\u001B[1;32m 262\u001B[0m \u001B[38;5;66;03m# #############################\u001B[39;00m\n\u001B[1;32m 263\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39msink_csv_figures()\n",
"File \u001B[0;32m~/Dropbox/GeneralCareerDevelop/PersonalBrand/GithubRepositories/ProSelfLC_tpami2release/src/proselflc/trainer/trainer_cnn_vision_derivedgrad.py:421\u001B[0m, in \u001B[0;36mTrainer.train_one_epoch\u001B[0;34m(self, epoch, dataloader)\u001B[0m\n\u001B[1;32m 408\u001B[0m \u001B[38;5;66;03m# print(str(logits.requires_grad))\u001B[39;00m\n\u001B[1;32m 409\u001B[0m \u001B[38;5;66;03m# print(logits.shape)\u001B[39;00m\n\u001B[1;32m 410\u001B[0m \u001B[38;5;66;03m# print(pred_probs.shape)\u001B[39;00m\n\u001B[0;32m (...)\u001B[0m\n\u001B[1;32m 418\u001B[0m \n\u001B[1;32m 419\u001B[0m \u001B[38;5;66;03m# for logging, I am still using epoch, only to reduce changes here.\u001B[39;00m\n\u001B[1;32m 420\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mcur_time \u001B[38;5;241m%\u001B[39m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mparams[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124meval_interval\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m==\u001B[39m \u001B[38;5;241m0\u001B[39m:\n\u001B[0;32m--> 421\u001B[0m \u001B[38;5;28;43mself\u001B[39;49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43meval_helper\u001B[49m\u001B[43m(\u001B[49m\u001B[43mepoch\u001B[49m\u001B[43m)\u001B[49m\n\u001B[1;32m 422\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mnetwork\u001B[38;5;241m.\u001B[39mtrain() \u001B[38;5;66;03m# self.network.train(mode=True)\u001B[39;00m\n\u001B[1;32m 424\u001B[0m \u001B[38;5;66;03m# update params\u001B[39;00m\n",
"File \u001B[0;32m~/Dropbox/GeneralCareerDevelop/PersonalBrand/GithubRepositories/ProSelfLC_tpami2release/src/proselflc/trainer/trainer_cnn_vision_derivedgrad.py:479\u001B[0m, in \u001B[0;36mTrainer.eval_helper\u001B[0;34m(self, epoch)\u001B[0m\n\u001B[1;32m 476\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mmax_p_dynamics[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mepoch\u001B[39m\u001B[38;5;124m\"\u001B[39m]\u001B[38;5;241m.\u001B[39mappend(\u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mcur_time)\n\u001B[1;32m 478\u001B[0m \u001B[38;5;28;01mfor\u001B[39;00m eval_dataname, eval_dataloader \u001B[38;5;129;01min\u001B[39;00m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mdataloaders\u001B[38;5;241m.\u001B[39mitems():\n\u001B[0;32m--> 479\u001B[0m (eval_loss, eval_accuracy, eval_entropy, eval_max_p) \u001B[38;5;241m=\u001B[39m \u001B[38;5;28;43mself\u001B[39;49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mevaluation\u001B[49m\u001B[43m(\u001B[49m\n\u001B[1;32m 480\u001B[0m \u001B[43m \u001B[49m\u001B[43mdataloader\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43meval_dataloader\u001B[49m\u001B[43m,\u001B[49m\n\u001B[1;32m 481\u001B[0m \u001B[43m \u001B[49m\u001B[43m)\u001B[49m\n\u001B[1;32m 482\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mloss_dynamics[eval_dataname]\u001B[38;5;241m.\u001B[39mappend(eval_loss)\n\u001B[1;32m 483\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39maccuracy_dynamics[eval_dataname]\u001B[38;5;241m.\u001B[39mappend(eval_accuracy)\n",
"File \u001B[0;32m~/.local/share/virtualenvs/ProSelfLC_tpami2release-X5r1NEDs/lib/python3.8/site-packages/torch/autograd/grad_mode.py:28\u001B[0m, in \u001B[0;36m_DecoratorContextManager.__call__.<locals>.decorate_context\u001B[0;34m(*args, **kwargs)\u001B[0m\n\u001B[1;32m 25\u001B[0m \u001B[38;5;129m@functools\u001B[39m\u001B[38;5;241m.\u001B[39mwraps(func)\n\u001B[1;32m 26\u001B[0m \u001B[38;5;28;01mdef\u001B[39;00m \u001B[38;5;21mdecorate_context\u001B[39m(\u001B[38;5;241m*\u001B[39margs, \u001B[38;5;241m*\u001B[39m\u001B[38;5;241m*\u001B[39mkwargs):\n\u001B[1;32m 27\u001B[0m \u001B[38;5;28;01mwith\u001B[39;00m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m\u001B[38;5;18m__class__\u001B[39m():\n\u001B[0;32m---> 28\u001B[0m \u001B[38;5;28;01mreturn\u001B[39;00m \u001B[43mfunc\u001B[49m\u001B[43m(\u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43margs\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mkwargs\u001B[49m\u001B[43m)\u001B[49m\n",
"File \u001B[0;32m~/Dropbox/GeneralCareerDevelop/PersonalBrand/GithubRepositories/ProSelfLC_tpami2release/src/proselflc/trainer/trainer_cnn_vision_derivedgrad.py:529\u001B[0m, in \u001B[0;36mTrainer.evaluation\u001B[0;34m(self, dataloader)\u001B[0m\n\u001B[1;32m 523\u001B[0m loss \u001B[38;5;241m=\u001B[39m cross_entropy(\n\u001B[1;32m 524\u001B[0m pred_probs\u001B[38;5;241m=\u001B[39mpred_probs,\n\u001B[1;32m 525\u001B[0m target_probs\u001B[38;5;241m=\u001B[39mlabels,\n\u001B[1;32m 526\u001B[0m )\n\u001B[1;32m 527\u001B[0m \u001B[38;5;66;03m# #############################\u001B[39;00m\n\u001B[0;32m--> 529\u001B[0m test_loss \u001B[38;5;241m+\u001B[39m\u001B[38;5;241m=\u001B[39m \u001B[43mloss\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mitem\u001B[49m\u001B[43m(\u001B[49m\u001B[43m)\u001B[49m\n\u001B[1;32m 530\u001B[0m _, preds \u001B[38;5;241m=\u001B[39m pred_probs\u001B[38;5;241m.\u001B[39mmax(\u001B[38;5;241m1\u001B[39m)\n\u001B[1;32m 531\u001B[0m _, annotations \u001B[38;5;241m=\u001B[39m labels\u001B[38;5;241m.\u001B[39mmax(\u001B[38;5;241m1\u001B[39m)\n",
"\u001B[0;31mKeyboardInterrupt\u001B[0m: "
]
}
],
"source": [
"trainer.train()\n",
"torch.save(\n",
" trainer.network,\n",
" params[\"summary_writer_dir\"] + \"/model.pt\",\n",
")\n",
"print(\n",
" \"The experiment is finished with details sinked in {}\".format(\n",
" params[\"summary_writer_dir\"]\n",
" )\n",
")"
],
"metadata": {
"collapsed": false
}
},
{
"cell_type": "code",
"execution_count": null,
"outputs": [],
"source": [],
"metadata": {
"collapsed": false
}
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"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": 1
}

0 comments on commit ea71596

Please sign in to comment.