image emb circle loss.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## DeiT"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 30/70 split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from imports import *\n",
    "from utils import load_data\n",
    "from image_train.data_with_timm_tfms import create_dl, ImageDS\n",
    "from image_train.model import EMBRes\n",
    "from image_train.train_with_circle_loss import *\n",
    "from timm.data import create_transform\n",
    "import matplotlib.pyplot as plt\n",
    "np.random.seed(1337)\n",
    "device = torch.device('cuda')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "df, train_df, val_df, train_labels, val_labels = load_data(df_path = '/home/dlo/Documents/product_match/data/train.csv',\n",
    "                                                           train_perc=0.3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_tfms = create_transform(\n",
    "            input_size=224,\n",
    "            is_training=True,\n",
    "            color_jitter=0.4,\n",
    "            auto_augment='rand-m9-mstd0.5-inc1',\n",
    "            interpolation='bicubic',\n",
    "            re_prob=0.25,\n",
    "            re_mode='pixel',\n",
    "            re_count=1,\n",
    "        )\n",
    "val_tfms = create_transform(\n",
    "            input_size=224,\n",
    "            is_training=False,\n",
    "            color_jitter=0.4,\n",
    "            auto_augment='rand-m9-mstd0.5-inc1',\n",
    "            interpolation='bicubic',\n",
    "            re_prob=0.25,\n",
    "            re_mode='pixel',\n",
    "            re_count=1,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# creating dataloaders\n",
    "small_images_dir_train = '/home/dlo/Documents/product_match/data/small_train_images/'\n",
    "small_images_dir_val = '/home/dlo/Documents/product_match/data/small_train_images/'\n",
    "bs = 256\n",
    "\n",
    "tr_dl = create_dl(train_df, small_images_dir_train, batch_size=bs, trans = train_tfms)\n",
    "tr_test_dl = create_dl(train_df, small_images_dir_val, shuffle=False, batch_size=bs, trans = train_tfms)\n",
    "val_dl = create_dl(val_df, small_images_dir_val, shuffle=False, batch_size=bs*2, trans = val_tfms)\n",
    "#full_dl = create_dl(df, small_images_dir, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "vision_model = 'efficientnet_b0'\n",
    "model = timm.create_model(vision_model, pretrained=True, num_classes=0).to('cuda')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class CircleLoss(nn.Module):\n",
    "    def __init__(self, m, gamma):\n",
    "        super(CircleLoss, self).__init__()\n",
    "        self.m = m\n",
    "        self.gamma = gamma\n",
    "        self.soft_plus = nn.Softplus()\n",
    "\n",
    "    def forward(self, sp, sn):\n",
    "        ap = torch.clamp_min(- sp.detach() + 1 + self.m, min=0.)\n",
    "        an = torch.clamp_min(sn.detach() + self.m, min=0.)\n",
    "\n",
    "        delta_p = 1 - self.m\n",
    "        delta_n = self.m\n",
    "\n",
    "        logit_p = - ap * (sp - delta_p) * self.gamma\n",
    "        logit_n = an * (sn - delta_n) * self.gamma\n",
    "\n",
    "        loss = self.soft_plus(torch.logsumexp(logit_n, dim=0) + torch.logsumexp(logit_p, dim=0))\n",
    "\n",
    "        return loss\n",
    "    \n",
    "metric_fc = CircleLoss(0.25, 256).to(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_epochs, lf, params, optimizer, sched = get_hparams(tr_dl, model, metric_fc, lr=1e-5, n_epochs=15)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss_hist = []\n",
    "best_thr_score={'val': {'thr': None, 'f1': None}, 'train': {'thr': None, 'f1': None}}\n",
    "thr_score_hist=[]\n",
    "ep_start = 0\n",
    "save_path = 'data/image_models/deit/test_5ap'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "cf65ccb0f93547c8a4fee4a077c5f455",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/15 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/41 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Thresholds:   0%|          | 0/10 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Checkpoint : saved model to data/image_models/deit/test_5ap_ep_0.pth\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "04e259ddb3dd44f49a0cb473a82449c9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/47 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "best_thr_score, thr_score_hist, losses = train(model, optimizer, lf, sched, metric_fc, tr_dl, val_dl,\n",
    "                                               n_epochs, train_df, val_df, train_tfms, val_tfms, \n",
    "                                               save_path=save_path, \n",
    "                                               prev_best_info=best_thr_score, info_history=thr_score_hist,\n",
    "                                               ep_start=ep_start)\n",
    "loss_hist.append(losses)\n",
    "ep_start += n_epochs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.eval()\n",
    "with torch.no_grad():\n",
    "    pbar = tqdm(val_dl, leave=False)\n",
    "    embs = []\n",
    "    for imgs, _ in pbar:\n",
    "        imgs = val_transforms(imgs).to('cuda')\n",
    "        feature = model(imgs)\n",
    "        embs.append(feature).cpu()\n",
    "    embs = F.normalize(torch.cat(embs, 0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.save(metric_fc, 'data/image_models/arcmarg_14ep_0.3.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_hist(history) : \n",
    "    train_scores = [info['train']['f1'] for info in history]\n",
    "    train_thr = [info['train']['thr'] for info in history]\n",
    "    val_scores = [info['val']['f1'] for info in history]\n",
    "    val_thr = [info['val']['thr'] for info in history]\n",
    "    x = range(len(train_scores))\n",
    "    plt.plot(x, val_scores, label='val_score')\n",
    "    plt.plot(x, train_scores, label='train_score')\n",
    "    plt.legend()\n",
    "    plt.show()\n",
    "    plt.plot(x, val_thr, label='val_thr')\n",
    "    plt.plot(x, train_thr, label='train_thr')\n",
    "    plt.legend()\n",
    "    plt.show()"
   ]
  }
 ],
 "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": "3.7.6"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}