smartpool-examples 0.1.3__tar.gz → 0.1.6__tar.gz

{smartpool_examples-0.1.3/smartpool_examples.egg-info → smartpool_examples-0.1.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: smartpool-examples
-Version: 0.1.3
+Version: 0.1.6
 Summary: Examples for smartpool.
 Author-email: "王炳辉 (Bing-Hui WANG)" <binghui.wang@foxmail.com>
 License: MIT
@@ -11,14 +11,14 @@ Classifier: License :: OSI Approved :: MIT License
 Classifier: Operating System :: OS Independent
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: pysmartpool
+Requires-Dist: pysmartpool>=0.1.6
 Requires-Dist: matplotlib
 Requires-Dist: scikit-learn
 Requires-Dist: numpy
 Requires-Dist: rich
-Requires-Dist: click
+Requires-Dist: typer
 Requires-Dist: joblib
-Requires-Dist: datawatcher
+Requires-Dist: opencv-python
 Dynamic: license-file
 
 # SmartPool Examples
@@ -79,6 +79,26 @@ python -m smartpool_examples.cross_validation --pool ray
 - Performance comparison with external frameworks
 
 
+### 3. ONNX Inference (`onnx_infer`)
+
+Runs batched ONNX model inference using `InferSessionPool` for concurrent GPU/CPU execution.
+Automatically manages inference sessions across worker threads.
+
+#### Running the Example
+
+```bash
+python -m smartpool_examples.onnx_infer --max-workers 4
+```
+
+#### What it Demonstrates
+
+- `InferSessionPool` creation and session lifecycle management
+- Multi-threaded inference with automatic device placement
+- COCO-format image preprocessing (resize, normalize, letterbox)
+- Softmax + top-5 postprocessing
+- Progress bars for downloads and inference steps
+
+
 ## License
 
 MIT License - see main smartpool repository for details

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/README.md

@@ -56,6 +56,26 @@ python -m smartpool_examples.cross_validation --pool ray
 - Performance comparison with external frameworks
 
 
+### 3. ONNX Inference (`onnx_infer`)
+
+Runs batched ONNX model inference using `InferSessionPool` for concurrent GPU/CPU execution.
+Automatically manages inference sessions across worker threads.
+
+#### Running the Example
+
+```bash
+python -m smartpool_examples.onnx_infer --max-workers 4
+```
+
+#### What it Demonstrates
+
+- `InferSessionPool` creation and session lifecycle management
+- Multi-threaded inference with automatic device placement
+- COCO-format image preprocessing (resize, normalize, letterbox)
+- Softmax + top-5 postprocessing
+- Progress bars for downloads and inference steps
+
+
 ## License
 
 MIT License - see main smartpool repository for details

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/pyproject.toml

@@ -1,10 +1,10 @@
 [build-system]
-requires = ["setuptools>=45", "wheel"]
+requires = ["setuptools", "wheel"]
 build-backend = "setuptools.build_meta"
 
 [project]
 name = "smartpool-examples"
-version = "0.1.3"
+version = "0.1.6"
 description = "Examples for smartpool."
 readme = "README.md"
 authors = [
@@ -17,16 +17,19 @@ classifiers = [
     "Operating System :: OS Independent",
 ]
 dependencies = [
-    "pysmartpool",
+    "pysmartpool>=0.1.6",
     "matplotlib",
     "scikit-learn",
     "numpy",
     "rich",
-    "click",
+    "typer",
     "joblib",
-    "datawatcher"
+    "opencv-python",
 ]
 
 [project.urls]
 Homepage = "https://github.com/Time-Coder/smartpool"
-Repository = "https://github.com/Time-Coder/smartpool.git"
+Repository = "https://github.com/Time-Coder/smartpool.git"
+
+[tool.setuptools.packages.find]
+include = ["smartpool_examples", "smartpool_examples.*"]

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/smartpool_examples/count_prime/__main__.py

@@ -4,28 +4,30 @@ import sys
 self_folder = os.path.dirname(os.path.abspath(__file__)).replace("\\", "/")
 sys.path.append(self_folder)
 
-from smartpool import ProcessPool
+if __name__ == "__main__":
+    target_folder = os.path.abspath(self_folder + "/../../../smartpool").replace("\\", "/")
+    sys.path.append(target_folder)
 
 from count_prime import count_prime
 
+from smartpool import ProcessPool
 
 if __name__ == "__main__":
     print("Use ProcessPool to count prime numbers lower than 10000.")
-    print(f"See source code at folder {os.path.dirname(os.path.abspath(__file__))}")
-    
+    print(f"See source code at folder {self_folder}")
+
     tasks = []
     start = 0
     while start < 10000:
         stop = start + 1000
         tasks.append((start, stop))
         start = stop
-        
+
     with ProcessPool() as pool:
         futures = []
         for task in tasks:
             future = pool.submit(count_prime, args=task)
             futures.append(future)
-        
+
         total_primes_count = sum(future.result() for future in futures)
         print(total_primes_count)
-    

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/smartpool_examples/count_prime/count_prime.py

@@ -4,10 +4,8 @@ import math
 def is_prime(num:int):
     if num < 2:
         return False
-    for i in range(2, int(math.sqrt(num)) + 1):
-        if num % i == 0:
-            return False
-    return True
+
+    return all(num % i != 0 for i in range(2, int(math.sqrt(num)) + 1))
 
 
 def count_prime(start:int, stop:int):
@@ -15,4 +13,4 @@ def count_prime(start:int, stop:int):
     for i in range(start, stop):
         if is_prime(i):
             count += 1
-    return count
+    return count

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/smartpool_examples/cross_validation/__main__.py

@@ -1,36 +1,61 @@
-from smartpool import ProcessPool, ThreadPool, DataSize, limit_num_single_thread
-limit_num_single_thread()
+if __name__ == "__main__":
+    import os
+    import sys
+    self_folder = os.path.dirname(os.path.abspath(__file__)).replace("\\", "/")
+    target_folder = os.path.abspath(self_folder + "/../../../smartpool").replace("\\", "/")
+    sys.path.append(target_folder)
 
-import click
-
-
-@click.command(help=f"Use smartpool to do 5-fold cross validatation for 7 deep learning models for handwritten digit recognition task.")
-@click.option(
-    '--pool', default='smartpool.ProcessPool', type=click.Choice([
-        'smartpool.ProcessPool',
-        'smartpool.ThreadPool',
-        'multiprocessing.Pool',
-        'concurrent.futures.ProcessPoolExecutor',
-        'concurrent.futures.ThreadPoolExecutor',
-        "joblib.Parallel(backend='loky')",
-        "joblib.Parallel(backend='threading')",
-        "ray"
-    ]),
-    help="choose process pool implementations"
-)
-@click.option(
-    '--max_workers', default=0, type=int,
-    help='max number of workers to use, 0 to use all available cores'
+
+from smartpool import (
+    DataSize,
+    ProcessPool,
+    Resource,
+    ThreadPool,
+    limit_num_single_thread,
 )
-def main(pool:str="smart", max_workers:int=0):
+
+limit_num_single_thread()
+
+from typing import Literal, TypeAlias
+
+import typer
+
+app = typer.Typer(help="Use smartpool to do 5-fold cross validatation for 7 deep learning models for handwritten digit recognition task.")
+
+PoolChoice: TypeAlias = Literal[
+    "smartpool.ProcessPool",
+    "smartpool.ThreadPool",
+    "multiprocessing.Pool",
+    "concurrent.futures.ProcessPoolExecutor",
+    "concurrent.futures.ThreadPoolExecutor",
+    "joblib.Parallel(backend='loky')",
+    "joblib.Parallel(backend='threading')",
+    "ray"
+]
+
+
+@app.command()
+def main(
+    pool: PoolChoice = typer.Option(
+        "smartpool.ProcessPool",
+        "--pool",
+        help="choose process pool implementations"
+    ),
+    max_workers: int = typer.Option(
+        0,
+        "--max_workers",
+        help="max number of workers to use, 0 to use all available cores"
+    )
+):
     import os
+    import importlib
     os.environ["RAY_ACCEL_ENV_VAR_OVERRIDE_ON_ZERO"] = "0"
-    
+
     print(f"Use {pool} to do 5-fold cross validatation for 7 deep learning models for handwritten digit recognition task.")
     print("Use `python -m smartpool_examples.cross_validation --help` to see all options.")
     print(f"See source code at folder {os.path.dirname(os.path.abspath(__file__))}")
     print("\npreparing data...")
-    
+
     try:
         import torch
         import torch.nn as nn
@@ -38,33 +63,31 @@ def main(pool:str="smart", max_workers:int=0):
         print("PyTorch is not installed. Follow https://pytorch.org/ instructions to install PyTorch.")
         exit(1)
 
-    try:
-        import torchvision
-    except ImportError:
+    if importlib.util.find_spec("torchvision") is None:
         print("torchvision is not installed. Use `pip install torchvision` to install torchvision.")
         exit(1)
 
-    import time
-    from sklearn.model_selection import KFold
-    import numpy as np
-    from rich.progress import Progress, BarColumn, TextColumn, TimeRemainingColumn
     import multiprocessing as mp
+    import os
     import queue
+    import sys
+    import time
     from collections import defaultdict
     from concurrent.futures import Future
     from typing import Dict, Union
 
-    import os
-    import sys
+    import numpy as np
+    from rich.progress import BarColumn, Progress, TextColumn, TimeRemainingColumn
+    from sklearn.model_selection import KFold
 
     self_folder = os.path.dirname(os.path.abspath(__file__)).replace("\\", "/")
     sys.path.append(self_folder)
 
     import models
+    from config import EPOCHS
     from data_utils import prepare_data
-    from model_utils import train_single_fold, ErrorInfo, ProgressInfo, TrainingResult
+    from model_utils import ErrorInfo, ProgressInfo, TrainingResult, train_single_fold
     from visualization import plot_results, print_results_table
-    from config import EPOCHS
 
     if max_workers == 0:
         max_workers = os.cpu_count()
@@ -73,10 +96,10 @@ def main(pool:str="smart", max_workers:int=0):
         cls for cls in models.__dict__.values()
         if isinstance(cls, type) and issubclass(cls, nn.Module) and cls != nn.Module
     ]
-    
+
     dataset = prepare_data()
     kfold = KFold(n_splits=5, shuffle=True, random_state=42)
-    
+
     manager = mp.Manager()
 
     if pool != "ray":
@@ -88,17 +111,17 @@ def main(pool:str="smart", max_workers:int=0):
         except ImportError:
             print("Ray is not installed. Use `pip install ray` to install Ray.")
             exit(1)
-        
+
         progress_queue:queue.Queue[Union[ProgressInfo, ErrorInfo]] = ray.util.queue.Queue()
 
     tasks = []
     for fold_idx, (train_indices, val_indices) in enumerate(kfold.split(dataset)):
         for model_class in model_classes:
             tasks.append((fold_idx, model_class, train_indices.copy(), val_indices.copy(), dataset, progress_queue))
-    
+
     task_progress_bars = {}
     best_device = 'cuda' if torch.cuda.is_available() else 'cpu'
-    
+
     start_time = time.perf_counter()
     with Progress(
         TextColumn("[progress.description]{task.description}"),
@@ -106,7 +129,7 @@ def main(pool:str="smart", max_workers:int=0):
         TextColumn("[progress.percentage]{task.percentage:>3.0f}%"),
         TimeRemainingColumn()
     ) as progress:
-        
+
         active_tasks = {}
 
         if pool == "smartpool.ProcessPool":
@@ -128,7 +151,7 @@ def main(pool:str="smart", max_workers:int=0):
         elif pool == "joblib.Parallel(backend='threading')":
             from joblib import Parallel, delayed
             process_pool = Parallel(n_jobs=max_workers, backend='threading', return_as="generator")
-        
+
         print("submitting training tasks...")
         futures_map:Dict[str, Future] = {}
         futures = []
@@ -137,10 +160,16 @@ def main(pool:str="smart", max_workers:int=0):
                 future = process_pool.submit(
                     train_single_fold,
                     args=task_args,
-                    need_cpu_cores=1,
-                    need_cpu_mem=1.1*DataSize.GB,
-                    need_gpu_cores=1000,
-                    need_gpu_mem=0.2*DataSize.GB
+                    cpu_mode_res=Resource(
+                        cpu_cores=1,
+                        cpu_mem=1.1*DataSize.GB
+                    ),
+                    gpu_mode_res=Resource(
+                        cpu_cores=1,
+                        cpu_mem=1.1*DataSize.GB,
+                        gpu_cores=1000,
+                        gpu_mem=0.2*DataSize.GB
+                    )
                 )
             elif pool.startswith("concurrent.futures."):
                 future = process_pool.submit(train_single_fold, *task_args, best_device if i % max_workers < 5 else 'cpu')
@@ -150,14 +179,14 @@ def main(pool:str="smart", max_workers:int=0):
                 future = delayed(train_single_fold)(*task_args, best_device if i % max_workers < 5 else 'cpu')
             elif pool == "ray":
                 future = ray.remote(num_cpus=1, num_gpus=(0.2 if i % max_workers < 5 else 0), memory=1.1*DataSize.GB)(train_single_fold).remote(*task_args, best_device if i % max_workers < 5 else 'cpu')
-            
+
             fold_idx = task_args[0]
             model_class = task_args[1]
             model_name = model_class.__name__
             task_key = f"{model_name}_fold_{fold_idx}"
             futures_map[task_key] = future
             futures.append(future)
-        
+
         print(f"training all models in {pool} ...")
         if pool.startswith("joblib"):
             joblib_results = process_pool(futures)
@@ -170,31 +199,31 @@ def main(pool:str="smart", max_workers:int=0):
                 break
 
             task_key = f"{progress_info.model_name}_fold_{progress_info.fold_idx}"
-            
+
             if task_key not in task_progress_bars:
                 initial_desc = f"train {progress_info.model_name} on {progress_info.device} "
                 initial_desc += f"for fold {progress_info.fold_idx+1}/5"
                 task_progress_bars[task_key] = progress.add_task(initial_desc, total=100)
                 active_tasks[task_key] = True
-            
+
             if task_key in task_progress_bars:
                 epoch_progress = (progress_info.epoch - 1) / 5
                 batch_progress = progress_info.batch / progress_info.total_batches
                 total_progress = (epoch_progress + batch_progress / 5) * 100
-                
+
                 if progress_info.epoch == 5 and progress_info.batch == progress_info.total_batches:
                     total_progress = 100.0
                     finished_tasks.add(task_key)
-                
+
                 new_desc = f"train {progress_info.model_name} on {progress_info.device} "
                 new_desc += f"for fold {progress_info.fold_idx+1}/5 - Epoch {progress_info.epoch}/{EPOCHS} "
                 new_desc += f"Loss: {progress_info.avg_loss:.4f} "
                 new_desc += f"Val Acc: {progress_info.val_accuracy*100:.2f}%"
                 if progress_info.device.startswith("cuda"):
                     new_desc = "[bright_cyan]" + new_desc
-                
+
                 progress.update(
-                    task_progress_bars[task_key], 
+                    task_progress_bars[task_key],
                     completed=total_progress,
                     description=new_desc
                 )
@@ -206,7 +235,7 @@ def main(pool:str="smart", max_workers:int=0):
 
     model_results = defaultdict(list)
     if pool in ["smartpool.ProcessPool", "smartpool.ThreadPool", "concurrent.futures.ProcessPoolExecutor", "concurrent.futures.ThreadPoolExecutor", "multiprocessing.Pool"]:
-        for task_key, future in futures_map.items():
+        for future in futures_map.values():
             if pool == "multiprocessing.Pool":
                 result:TrainingResult = future.get()
             else:
@@ -220,7 +249,7 @@ def main(pool:str="smart", max_workers:int=0):
         ray_results = ray.get(futures)
         for result in ray_results:
             model_results[result.model_name].append(result.val_accuracy)
-    
+
     stop_time = time.perf_counter()
     print(f"train completed in {stop_time - start_time:.2f} seconds")
 
@@ -235,10 +264,10 @@ def main(pool:str="smart", max_workers:int=0):
             'max': np.max(accuracies),
             'accuracies': accuracies
         }
-    
+
     print_results_table(stats)
     plot_results(model_results, stats)
 
 
 if __name__ == "__main__":
-    main()
+    app()

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/smartpool_examples/cross_validation/config.py

@@ -1,11 +1,12 @@
+import os
+
+self_folder = os.path.dirname(os.path.abspath(__file__))
+
 # Training parameters
 BATCH_SIZE = 128
 EPOCHS = 5
 LEARNING_RATE = 0.001
 
 # Data settings
-import os
-self_folder = os.path.dirname(os.path.abspath(__file__))
-
 DATA_ROOT = f'{self_folder}/data'
 DATASET_NAME = 'MNIST'

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/smartpool_examples/cross_validation/data_utils.py

@@ -1,25 +1,25 @@
 import os
+
+from config import BATCH_SIZE, DATA_ROOT
 from torch.utils.data import DataLoader, Subset
 from torchvision import datasets, transforms
 
-from config import DATA_ROOT, BATCH_SIZE
-
 
 def prepare_data():
     transform = transforms.Compose([
         transforms.ToTensor(),
         transforms.Normalize((0.1307,), (0.3081,))
     ])
-    
+
     mnist_exists = (
         os.path.exists(os.path.join(DATA_ROOT, 'MNIST', 'raw')) and
         os.path.exists(os.path.join(DATA_ROOT, 'MNIST', 'processed'))
     )
-    
+
     dataset = datasets.MNIST(
-        root=DATA_ROOT, 
-        train=True, 
-        download=not mnist_exists, 
+        root=DATA_ROOT,
+        train=True,
+        download=not mnist_exists,
         transform=transform
     )
     dataset.data.share_memory_()
@@ -30,18 +30,18 @@ def prepare_data():
 def create_data_loaders(dataset, train_indices, val_indices):
     train_subset = Subset(dataset, train_indices)
     val_subset = Subset(dataset, val_indices)
-    
+
     train_loader = DataLoader(
-        train_subset, 
-        batch_size=BATCH_SIZE, 
-        shuffle=True, 
+        train_subset,
+        batch_size=BATCH_SIZE,
+        shuffle=True,
         pin_memory=True
     )
     val_loader = DataLoader(
-        val_subset, 
-        batch_size=BATCH_SIZE, 
-        shuffle=False, 
+        val_subset,
+        batch_size=BATCH_SIZE,
+        shuffle=False,
         pin_memory=True
     )
-    
-    return train_loader, val_loader
+
+    return train_loader, val_loader

{smartpool_examples-0.1.3 → smartpool_examples-0.1.6}/smartpool_examples/cross_validation/model_utils.py

@@ -1,14 +1,13 @@
-import torch
-import torch.nn as nn
-import torch.optim as optim
-
 import traceback
 from dataclasses import dataclass
 
-from config import LEARNING_RATE, EPOCHS
+import torch
+import torch.nn as nn
+import torch.optim as optim
+from config import EPOCHS, LEARNING_RATE
 from data_utils import create_data_loaders
 
-from smartpool import move_optimizer_to, best_device
+from smartpool import best_device, move_optimizer_to
 
 
 @dataclass
@@ -47,17 +46,17 @@ def _train_single_fold(fold_idx, model_class, train_indices, val_indices, datase
     train_loader, val_loader = create_data_loaders(dataset, train_indices, val_indices)
     num_batches = len(train_loader)
     model = model_class()
-    
+
     device = user_device
     if user_device is None:
         device = best_device()
 
     old_device = device
     model.to(device, non_blocking=True)
-    
+
     criterion = nn.CrossEntropyLoss()
     optimizer = optim.Adam(model.parameters(), lr=LEARNING_RATE)
-    
+
     initial_progress = ProgressInfo(
         model_name=model_class.__name__,
         fold_idx=fold_idx,
@@ -71,7 +70,7 @@ def _train_single_fold(fold_idx, model_class, train_indices, val_indices, datase
     progress_queue.put(initial_progress)
 
     last_val_accuracy = 0.0
-    
+
     for epoch in range(EPOCHS):
         epoch_loss = 0.0
         model.train()
@@ -81,20 +80,20 @@ def _train_single_fold(fold_idx, model_class, train_indices, val_indices, datase
 
             data = data.to(device, non_blocking=True)
             target = target.to(device, non_blocking=True)
-                    
+
             if device != old_device:
                 model.to(device, non_blocking=True)
                 move_optimizer_to(optimizer, device)
                 old_device = device
-                    
+
             optimizer.zero_grad()
             output = model(data)
             loss = criterion(output, target)
             loss.backward()
             optimizer.step()
-                    
+
             epoch_loss += loss.item()
-                    
+
             progress_info = ProgressInfo(
                 model_name=model_class.__name__,
                 fold_idx=fold_idx,
@@ -106,12 +105,12 @@ def _train_single_fold(fold_idx, model_class, train_indices, val_indices, datase
                 val_accuracy=last_val_accuracy
             )
             progress_queue.put(progress_info)
-        
+
         model.eval()
         correct = 0
         total = 0
         val_accuracy = 0.0
-        
+
         with torch.no_grad():
             for data, target in val_loader:
                 if user_device is None:
@@ -119,21 +118,21 @@ def _train_single_fold(fold_idx, model_class, train_indices, val_indices, datase
 
                 data = data.to(device, non_blocking=True)
                 target = target.to(device, non_blocking=True)
-                        
+
                 if device != old_device:
                     model.to(device, non_blocking=True)
                     move_optimizer_to(optimizer, device)
                     old_device = device
-                        
+
                 output = model(data)
                 pred = output.argmax(dim=1, keepdim=True)
                 correct += pred.eq(target.view_as(pred)).sum().item()
                 total += target.size(0)
-        
+
         val_accuracy = correct / total
         last_val_accuracy = val_accuracy
         model.train()
-        
+
         final_progress = ProgressInfo(
             model_name=model_class.__name__,
             fold_idx=fold_idx,
@@ -145,11 +144,11 @@ def _train_single_fold(fold_idx, model_class, train_indices, val_indices, datase
             val_accuracy=val_accuracy
         )
         progress_queue.put(final_progress)
-    
+
     model.eval()
     correct = 0
     total = 0
-    
+
     with torch.no_grad():
         for data, target in val_loader:
             if user_device is None:
@@ -167,7 +166,7 @@ def _train_single_fold(fold_idx, model_class, train_indices, val_indices, datase
             pred = output.argmax(dim=1, keepdim=True)
             correct += pred.eq(target.view_as(pred)).sum().item()
             total += target.size(0)
-    
+
     val_accuracy = correct / total
-    
+
     return TrainingResult(fold_idx, model_class.__name__, val_accuracy)