Update setup files

Makefile

@@ -1,27 +1,14 @@
-STAGED := $(shell git diff --cached --name-only --diff-filter=ACMR -- 'src/***.py' | sed 's| |\\ |g')
-
-all: format lint
- echo 'Makefile for meta-learning-for-everyone repository'
+all: init format lint
 
 format:
- black .
+ black . --line-length 104
  isort .
  nbqa black .
  nbqa isort .
 
 lint:
  pytest src/ --pylint --flake8 --ignore=src/meta_rl/envs
 
-lint-all:
- pytest src/ --pylint --flake8 --ignore=src/meta_rl/envs --cache-clear
-
-lint-staged:
-ifdef STAGED
- pytest $(STAGED) --pylint --flake8 --ignore=src/meta_rl/envs --cache-clear
-else
- @echo "No Staged Python File in the src folder"
-endif
-
 init:
  pip install -U pip
  pip install -e .

requirements.txt

@@ -7,5 +7,7 @@ jupyter-contrib-nbextensions==0.5.1
 jupyter-nbextensions-configurator==0.4.1
 matplotlib>=3.5.2
 mujoco==2.2.0
+psutil==5.9.1
+setuptools==59.5.0
 torchmeta>=1.8.0
 tqdm>=4.62.3

scripts/download-torch.py

@@ -8,11 +8,11 @@
 def main() -> None:
  if sys.platform == "win32" or sys.platform == "linux":
  if GPUtil.getAvailable():
- cli = "pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html"
+ cli = "pip install torch==1.9.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html"
  else:
- cli = "pip install torch==1.8.1+cpu torchvision==0.9.1+cpu torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html"
+ cli = "pip install torch==1.9.1+cpu -f https://download.pytorch.org/whl/torch_stable.html"
  elif sys.platform == "darwin":
- cli = "pip install torch==1.8.1 torchvision==0.9.1 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html"
+ cli = "pip install torch==1.9.1 -f https://download.pytorch.org/whl/torch_stable.html"
  print(cli)
  os.system(cli)
 

setup.cfg

@@ -0,0 +1,4 @@
+[isort]
+line_length = 104
+skip_gitignore = true
+extend_skip_glob = ""

src/meta_sl/load_dataset/load_sinusoid.ipynb

@@ -34,7 +34,9 @@
  " num_tasks=config[\"num_batches_train\"] * config[\"batch_size\"],\n",
  " noise_std=None,\n",
  " )\n",
- " train_dataloader = BatchMetaDataLoader(train_dataset, batch_size=config[\"batch_size\"])\n",
+ " train_dataloader = BatchMetaDataLoader(\n",
+ " train_dataset, batch_size=config[\"batch_size\"]\n",
+ " )\n",
  "\n",
  " val_dataset = Sinusoid(\n",
  " num_samples_per_task=config[\"num_shots\"] * 2,\n",
@@ -94,10 +96,18 @@
  "source": [
  "for batch_idx, batch in enumerate(val_dataloader):\n",
  " xs, ys = batch\n",
- " support_xs = xs[:, : config[\"num_shots\"], :].to(device=config[\"device\"]).type(torch.float)\n",
- " query_xs = xs[:, config[\"num_shots\"] :, :].to(device=config[\"device\"]).type(torch.float)\n",
- " support_ys = ys[:, : config[\"num_shots\"], :].to(device=config[\"device\"]).type(torch.float)\n",
- " query_ys = ys[:, config[\"num_shots\"] :, :].to(device=config[\"device\"]).type(torch.float)\n",
+ " support_xs = (\n",
+ " xs[:, : config[\"num_shots\"], :].to(device=config[\"device\"]).type(torch.float)\n",
+ " )\n",
+ " query_xs = (\n",
+ " xs[:, config[\"num_shots\"] :, :].to(device=config[\"device\"]).type(torch.float)\n",
+ " )\n",
+ " support_ys = (\n",
+ " ys[:, : config[\"num_shots\"], :].to(device=config[\"device\"]).type(torch.float)\n",
+ " )\n",
+ " query_ys = (\n",
+ " ys[:, config[\"num_shots\"] :, :].to(device=config[\"device\"]).type(torch.float)\n",
+ " )\n",
  "\n",
  " print(\n",
  " f\"support_x shape : {support_xs.shape}\\n\",\n",

src/meta_sl/metric-based/matching_network.ipynb

@@ -108,10 +108,14 @@
  " batch_first=True,\n",
  " )\n",
  "\n",
- " self.lstm_cell = nn.LSTMCell(input_size=self.emb_size, hidden_size=self.emb_size)\n",
+ " self.lstm_cell = nn.LSTMCell(\n",
+ " input_size=self.emb_size, hidden_size=self.emb_size\n",
+ " )\n",
  "\n",
  " @classmethod\n",
- " def convBlock(cls, in_channels: int, out_channels: int, kernel_size: int) -> nn.Sequential:\n",
+ " def convBlock(\n",
+ " cls, in_channels: int, out_channels: int, kernel_size: int\n",
+ " ) -> nn.Sequential:\n",
  " return nn.Sequential(\n",
  " nn.Conv2d(in_channels, out_channels, kernel_size, padding=1),\n",
  " nn.BatchNorm2d(out_channels, momentum=1.0, track_running_stats=False),\n",
@@ -125,7 +129,9 @@
  " dim=0, sizes=[batch_size, self.num_support]\n",
  " )\n",
  " support_bilstm, _ = self.bilstm(support_cnn)\n",
- " support_bilstm_for, support_bilstm_rev = torch.tensor_split(support_bilstm, 2, dim=-1)\n",
+ " support_bilstm_for, support_bilstm_rev = torch.tensor_split(\n",
+ " support_bilstm, 2, dim=-1\n",
+ " )\n",
  " return support_bilstm_for + support_bilstm_rev + support_cnn\n",
  "\n",
  " def read_out(self, hidden: torch.Tensor, support_emb: torch.Tensor) -> torch.Tensor:\n",
@@ -150,11 +156,15 @@
  " cell_state = query_cnn.new_zeros(query_cnn.shape)\n",
  "\n",
  " for _ in range(10):\n",
- " hidden_state, cell_state = self.lstm_cell(query_cnn, (hidden_state + read_out, cell_state))\n",
+ " hidden_state, cell_state = self.lstm_cell(\n",
+ " query_cnn, (hidden_state + read_out, cell_state)\n",
+ " )\n",
  " hidden_state = hidden_state + query_cnn\n",
  " read_out = self.read_out(hidden_state, support_emb)\n",
  "\n",
- " query_emb = hidden_state.unflatten(dim=0, sizes=[query_x.shape[0], query_x.shape[1]])\n",
+ " query_emb = hidden_state.unflatten(\n",
+ " dim=0, sizes=[query_x.shape[0], query_x.shape[1]]\n",
+ " )\n",
  " return query_emb\n",
  "\n",
  " def forward(\n",
@@ -191,7 +201,9 @@
  " support_emb_repeat, query_emb_repeat, dim=-1, eps=1e-8\n",
  " ) # batch_size, num_query, num_support\n",
  " attention = F.softmax(similarity, dim=-1) # batch_size, num_query, num_support\n",
- " indices = support_y.unsqueeze(1).expand(-1, num_query, -1) # batch_size, num_query, num_support\n",
+ " indices = support_y.unsqueeze(1).expand(\n",
+ " -1, num_query, -1\n",
+ " ) # batch_size, num_query, num_support\n",
  "\n",
  " prob = attention.new_zeros((batch_size, num_query, num_ways))\n",
  " prob.scatter_add_(-1, indices, attention) # batch_size, num_query, num_ways\n",
@@ -392,9 +404,9 @@
  "\n",
  "train_dataloader, val_dataloader, test_dataloader = get_dataloader(config)\n",
  "\n",
- "model = MatchingNet(in_channels=1, num_ways=config[\"num_ways\"], num_shots=config[\"num_shots\"]).to(\n",
- " device=config[\"device\"]\n",
- ")\n",
+ "model = MatchingNet(\n",
+ " in_channels=1, num_ways=config[\"num_ways\"], num_shots=config[\"num_shots\"]\n",
+ ").to(device=config[\"device\"])\n",
  "\n",
  "criterion = nn.CrossEntropyLoss()\n",
  "optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)"
@@ -452,7 +464,9 @@
  ],
  "source": [
  "# 메타-트레이닝\n",
- "with tqdm(zip(train_dataloader, val_dataloader), total=config[\"num_task_batch_train\"]) as pbar:\n",
+ "with tqdm(\n",
+ " zip(train_dataloader, val_dataloader), total=config[\"num_task_batch_train\"]\n",
+ ") as pbar:\n",
  " train_accuracies, val_accuracies = [], []\n",
  " train_losses, val_losses = [], []\n",
  "\n",
@@ -489,7 +503,9 @@
  " )\n",
  "\n",
  " # 모델 저장하기\n",
- " save_model(output_folder=config[\"output_folder\"], model=model, title=\"matching_network.th\")\n",
+ " save_model(\n",
+ " output_folder=config[\"output_folder\"], model=model, title=\"matching_network.th\"\n",
+ " )\n",
  "\n",
  " print_graph(\n",
  " train_accuracies=train_accuracies,\n",
@@ -538,7 +554,9 @@
  ],
  "source": [
  "# 모델 불러오기\n",
- "load_model(output_folder=config[\"output_folder\"], model=model, title=\"matching_network.th\")\n",
+ "load_model(\n",
+ " output_folder=config[\"output_folder\"], model=model, title=\"matching_network.th\"\n",
+ ")\n",
  "\n",
  "# 메타-테스팅\n",
  "with tqdm(test_dataloader, total=config[\"num_task_batch_test\"]) as pbar:\n",

src/meta_sl/metric-based/prototypical_network.ipynb

@@ -102,36 +102,44 @@
  " )\n",
  "\n",
  " @classmethod\n",
- " def convBlock(cls, in_channels: int, out_channels: int, kernel_size: int) -> nn.Sequential:\n",
+ " def convBlock(\n",
+ " cls, in_channels: int, out_channels: int, kernel_size: int\n",
+ " ) -> nn.Sequential:\n",
  " return nn.Sequential(\n",
  " nn.Conv2d(in_channels, out_channels, kernel_size, padding=1),\n",
  " nn.BatchNorm2d(out_channels, momentum=1.0, track_running_stats=False),\n",
  " nn.ReLU(),\n",
  " nn.MaxPool2d(2),\n",
  " )\n",
  "\n",
- " def get_prototypes(self, embeddings: torch.Tensor, targets: torch.Tensor) -> torch.Tensor:\n",
+ " def get_prototypes(\n",
+ " self, embeddings: torch.Tensor, targets: torch.Tensor\n",
+ " ) -> torch.Tensor:\n",
  " batch_size = embeddings.shape[0]\n",
  " indices = targets.unsqueeze(-1).expand_as(embeddings)\n",
  "\n",
  " prototypes = embeddings.new_zeros((batch_size, self.num_ways, self.emb_size))\n",
- " prototypes.scatter_add_(1, indices, embeddings).div_(float(self.num_support) / self.num_ways)\n",
+ " prototypes.scatter_add_(1, indices, embeddings).div_(\n",
+ " float(self.num_support) / self.num_ways\n",
+ " )\n",
  " return prototypes\n",
  "\n",
  " def forward(\n",
  " self, support_x: torch.Tensor, support_y: torch.Tensor, query_x: torch.Tensor\n",
  " ) -> torch.Tensor:\n",
  " batch_size = support_x.shape[0]\n",
  "\n",
- " support_emb = self.embedding_net(support_x.flatten(start_dim=0, end_dim=1)).unflatten(\n",
- " dim=0, sizes=[batch_size, self.num_support]\n",
- " )\n",
- " query_emb = self.embedding_net(query_x.flatten(start_dim=0, end_dim=1)).unflatten(\n",
- " dim=0, sizes=[batch_size, self.num_query]\n",
- " )\n",
+ " support_emb = self.embedding_net(\n",
+ " support_x.flatten(start_dim=0, end_dim=1)\n",
+ " ).unflatten(dim=0, sizes=[batch_size, self.num_support])\n",
+ " query_emb = self.embedding_net(\n",
+ " query_x.flatten(start_dim=0, end_dim=1)\n",
+ " ).unflatten(dim=0, sizes=[batch_size, self.num_query])\n",
  " proto_emb = self.get_prototypes(support_emb, support_y)\n",
  "\n",
- " distance = torch.sum((query_emb.unsqueeze(2) - proto_emb.unsqueeze(1)) ** 2, dim=-1)\n",
+ " distance = torch.sum(\n",
+ " (query_emb.unsqueeze(2) - proto_emb.unsqueeze(1)) ** 2, dim=-1\n",
+ " )\n",
  " return distance"
  ]
  },
@@ -278,9 +286,9 @@
  "\n",
  "train_dataloader, val_dataloader, test_dataloader = get_dataloader(config)\n",
  "\n",
- "model = PrototypicalNet(in_channels=1, num_ways=config[\"num_ways\"], num_shots=config[\"num_shots\"]).to(\n",
- " device=config[\"device\"]\n",
- ")\n",
+ "model = PrototypicalNet(\n",
+ " in_channels=1, num_ways=config[\"num_ways\"], num_shots=config[\"num_shots\"]\n",
+ ").to(device=config[\"device\"])\n",
  "\n",
  "criterion = nn.CrossEntropyLoss()\n",
  "optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)"
@@ -338,7 +346,9 @@
  ],
  "source": [
  "# 메타-트레이닝\n",
- "with tqdm(zip(train_dataloader, val_dataloader), total=config[\"num_task_batch_train\"]) as pbar:\n",
+ "with tqdm(\n",
+ " zip(train_dataloader, val_dataloader), total=config[\"num_task_batch_train\"]\n",
+ ") as pbar:\n",
  " train_accuracies, val_accuracies = [], []\n",
  " train_losses, val_losses = [], []\n",
  "\n",
@@ -354,7 +364,10 @@
  " optimizer=optimizer,\n",
  " )\n",
  " val_accuracy, val_loss = test_proto(\n",
- " device=config[\"device\"], task_batch=val_batch, model=model, criterion=criterion\n",
+ " device=config[\"device\"],\n",
+ " task_batch=val_batch,\n",
+ " model=model,\n",
+ " criterion=criterion,\n",
  " )\n",
  "\n",
  " train_accuracies.append(train_accuracy)\n",
@@ -370,7 +383,11 @@
  " )\n",
  "\n",
  " # 모델 저장하기\n",
- " save_model(output_folder=config[\"output_folder\"], model=model, title=\"prototypical_network.th\")\n",
+ " save_model(\n",
+ " output_folder=config[\"output_folder\"],\n",
+ " model=model,\n",
+ " title=\"prototypical_network.th\",\n",
+ " )\n",
  "\n",
  " print_graph(\n",
  " train_accuracies=train_accuracies,\n",
@@ -427,7 +444,9 @@
  ],
  "source": [
  "# 모델 불러오기\n",
- "load_model(output_folder=config[\"output_folder\"], model=model, title=\"prototypical_network.th\")\n",
+ "load_model(\n",
+ " output_folder=config[\"output_folder\"], model=model, title=\"prototypical_network.th\"\n",
+ ")\n",
  "\n",
  "# 메타-테스팅\n",
  "with tqdm(test_dataloader, total=config[\"num_task_batch_test\"]) as pbar:\n",
@@ -439,7 +458,10 @@
  " break\n",
  "\n",
  " test_accuracy, test_loss = test_proto(\n",
- " device=config[\"device\"], task_batch=test_batch, model=model, criterion=criterion\n",
+ " device=config[\"device\"],\n",
+ " task_batch=test_batch,\n",
+ " model=model,\n",
+ " criterion=criterion,\n",
  " )\n",
  " sum_test_accuracies += test_accuracy\n",
  " sum_test_losses += test_loss\n",