tensorstudio 1.9.0__tar.gz → 1.11.0__tar.gz

{tensorstudio-1.9.0 → tensorstudio-1.11.0}/CHANGELOG.md

@@ -2,6 +2,44 @@
 
 ## Unreleased
 
+## 1.11.0 - 2026-07-07
+
+- Completed the ordered Computer Vision Depth roadmap section as one release
+  batch.
+- Added batch-aware resize/crop/normalize helpers plus color jitter, random
+  resized crop, random rotation, affine transforms, cutout, mixup, and CutMix.
+- Added detection utilities for box areas, IoU variants, NMS, box
+  encode/decode, coordinate conversion, and anchor generation.
+- Added segmentation helpers for mask IoU, one-hot conversion, masks-to-boxes,
+  nearest mask resize, and deterministic mask crops.
+- Added `DetectionFolder` and `SegmentationFolder` datasets plus
+  `tensorstudio.data` factory aliases for detection and segmentation folders.
+- Added `ResidualBlock`, `DepthwiseSeparableBlock`, `CompactUNet`, and
+  `make_unet()` vision model helpers using the native-backed neural-network
+  layer stack.
+- Added prediction drawing, mask overlay, and feature-map grid visualization
+  helpers.
+- Expanded vision tests and documentation for transforms, detection,
+  segmentation, model blocks, and visualization.
+
+## 1.10.0 - 2026-07-07
+
+- Completed the ordered Training And Project Workflows roadmap section as one
+  release batch.
+- Added `ArrayDataset`, tensor/array/image-folder dataset factories,
+  deterministic train/validation splitting, and dataset metadata summaries.
+- Added `tensorstudio.metrics` with regression, classification, and multilabel
+  metrics for small supervised workflows.
+- Added trainer validation loops, scheduler stepping, callback context support,
+  learning-rate logging, CSV logging, checkpoint callbacks, and early stopping.
+- Added JSON, TOML, and YAML project config loading plus deterministic seeding
+  across TensorStudio, NumPy, and Python random.
+- Expanded full checkpoints with scheduler and epoch state and added
+  `resume_checkpoint()` for continuing training runs.
+- Added generated regression, classification, and vision project templates.
+- Expanded tests and docs for project workflows, metrics, dataset creation,
+  callbacks, configs, templates, and checkpoint resume.
+
 ## 1.9.0 - 2026-07-07
 
 - Completed the ordered Neural Network Building Blocks roadmap section as one

{tensorstudio-1.9.0 → tensorstudio-1.11.0}/CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.18)
 
-project(tensorstudio VERSION 1.9.0 LANGUAGES CXX)
+project(tensorstudio VERSION 1.11.0 LANGUAGES CXX)
 
 find_package(Python COMPONENTS Interpreter Development.Module REQUIRED)
 set(PYBIND11_FINDPYTHON ON)

{tensorstudio-1.9.0 → tensorstudio-1.11.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: tensorstudio
-Version: 1.9.0
+Version: 1.11.0
 Summary: TensorStudio is a compact C++ tensor and autograd engine with a Python API for learning, experimentation, and lightweight ML workloads.
 Keywords: tensor,autograd,machine-learning,cpp,pybind11
 Author: TensorStudio contributors
@@ -42,6 +42,8 @@ Project-URL: Repository, https://github.com/imattas/TensorStudio
 Project-URL: Issues, https://github.com/imattas/TensorStudio/issues
 Requires-Python: >=3.10
 Requires-Dist: numpy>=1.24
+Requires-Dist: PyYAML>=6
+Requires-Dist: tomli>=2.0; python_version < "3.11"
 Provides-Extra: dev
 Requires-Dist: pytest>=8; extra == "dev"
 Requires-Dist: pytest-cov>=5; extra == "dev"
@@ -71,15 +73,18 @@ Description-Content-Type: text/markdown
 TensorStudio is a compact C++ tensor and autograd engine with a Python API for
 learning, experimentation, and lightweight ML workloads.
 
-TensorStudio `1.9.0` is a CPU-only stable API foundation with native C++
+TensorStudio `1.11.0` is a CPU-only stable API foundation with native C++
 threading, storage reuse, SIMD-friendly typed kernels, and optional
 CBLAS/Accelerate matrix multiplication when available. It adds native stable
 softmax/logsumexp, batched matrix multiplication, statistical reductions,
 boolean reductions, seeded random distributions, and a hardened eager autograd
 lifecycle. The neural-network layer now includes grouped/depthwise/1D/
 transposed convolution, normalization layers, embeddings, richer activations,
-initializers, additional losses, and model summaries. It is eager-only,
-intentionally compact, and not a replacement for mature ML frameworks.
+initializers, additional losses, and model summaries. The project layer adds
+dataset factories, deterministic train/validation splitting, metrics, callbacks,
+multi-format configs, checkpoint resume helpers, and starter project templates.
+It is eager-only, intentionally compact, and not a replacement for mature ML
+frameworks.
 
 ## Install
 
@@ -289,7 +294,7 @@ print(model.state_dict().keys())
 print(model.parameter_count())
 ```
 
-The `1.9.0` neural-network surface also includes initialization helpers,
+The neural-network surface also includes initialization helpers,
 normalization layers, embeddings, grouped/depthwise/1D/transposed convolution,
 adaptive/global pooling, richer activations, and model summaries:
 
@@ -313,6 +318,11 @@ classification workflows: Pillow-backed image IO, transform pipelines,
 deterministic augmentations, `ImageFolder` datasets, metrics, image grids,
 bounding-box drawing, and compact CNN classifiers running through native
 Conv2d/pooling kernels.
+The `1.11.0` vision surface adds batch-aware transforms, color jitter, random
+resized crop, rotation, affine transforms, cutout, mixup, CutMix, detection
+helpers, segmentation mask helpers, detection/segmentation folder datasets,
+ResNet-style blocks, MobileNet-style depthwise blocks, a compact UNet, and
+prediction/mask/feature-map visualization helpers.
 
 ```python
 import numpy as np
@@ -340,17 +350,22 @@ optimizer.step()
 print(ts.vision.accuracy(model(x), target))
 ```
 
-## DataLoader
+## Data And Metrics
 
 ```python
 import tensorstudio as ts
-from tensorstudio.data import DataLoader, TensorDataset
+from tensorstudio.data import DataLoader, from_arrays, train_val_split
 
-dataset = TensorDataset(ts.arange(6).reshape((6, 1)), ts.arange(6))
-loader = DataLoader(dataset, batch_size=2, shuffle=True, seed=42)
+dataset = from_arrays([[0.0], [1.0], [2.0], [3.0]], [[1.0], [3.0], [5.0], [7.0]])
+train_data, val_data = train_val_split(dataset, val_fraction=0.25, seed=42)
+loader = DataLoader(train_data, batch_size=2, shuffle=True, seed=42)
 
 for features, targets in loader:
     print(features.shape, targets.shape)
+
+prediction = ts.tensor([[0.2, 0.8], [0.7, 0.3]])
+target = ts.tensor([1, 0], dtype="int64")
+print(ts.metrics.accuracy(prediction, target))
 ```
 
 The v1 DataLoader is intentionally single-process so it works cleanly on
@@ -358,24 +373,44 @@ Windows without multiprocessing setup.
 
 ## Projects And Training
 
-`tensorstudio.project` provides project folders, JSON config, reusable trainers,
-safe NPZ weight files, and trusted full checkpoints:
+`tensorstudio.project` provides project folders, JSON/TOML/YAML config loading,
+deterministic seeding, reusable trainers, validation loops, callbacks, safe NPZ
+weight files, trusted full checkpoints, and generated starter templates:
 
 ```python
 import tensorstudio as ts
 from tensorstudio import nn, optim
-from tensorstudio.data import DataLoader, TensorDataset
-from tensorstudio.project import Project, ProjectConfig, Trainer, save_state_dict
+from tensorstudio.data import DataLoader, from_arrays, train_val_split
+from tensorstudio.project import (
+    CSVLogger,
+    CheckpointCallback,
+    LrLogger,
+    Project,
+    ProjectConfig,
+    Trainer,
+    save_state_dict,
+    seed_everything,
+)
 
-x = ts.tensor([[0.0], [1.0], [2.0], [3.0]])
-y = ts.tensor([[1.0], [3.0], [5.0], [7.0]])
+seed_everything(7)
+dataset = from_arrays([[0.0], [1.0], [2.0], [3.0]], [[1.0], [3.0], [5.0], [7.0]])
+train_data, val_data = train_val_split(dataset, val_fraction=0.25, seed=7)
 
 model = nn.Linear(1, 1)
-loader = DataLoader(TensorDataset(x, y), batch_size=2)
-trainer = Trainer(model, optim.SGD(model.parameters(), lr=0.05), nn.MSELoss())
+optimizer = optim.SGD(model.parameters(), lr=0.05)
+trainer = Trainer(model, optimizer, nn.MSELoss(), metric_fn=ts.metrics.mean_squared_error)
 project = Project("runs/linear", ProjectConfig(name="linear-regression", seed=7))
 
-history = trainer.fit(loader, epochs=50)
+history = trainer.fit(
+    DataLoader(train_data, batch_size=2),
+    epochs=50,
+    validation_loader=DataLoader(val_data, batch_size=1),
+    callbacks=[
+        LrLogger(),
+        CSVLogger(project.logs_dir / "history.csv"),
+        CheckpointCallback(project.checkpoints_dir / "best.tsmodel", save_best_only=True),
+    ],
+)
 save_state_dict(model, project.checkpoint_path("weights"))
 print(history.last)
 ```
@@ -410,11 +445,11 @@ Run the loose local regression thresholds with:
 python benchmark_all.py --check-thresholds
 ```
 
-On one Windows CPython 3.10 development run reporting `1.9.0`, with
+On one Windows CPython 3.10 development run reporting `1.11.0`, with
 TensorStudio threads enabled, storage pooling enabled, SSE2 autovectorization
 reported, and no BLAS provider found, TensorStudio beat NumPy on 7 local
 benchmark cases and lost on 96 NumPy-comparable cases. JAX CPU dispatch was
-available on that machine; TensorStudio won 55 local cases and lost 43. The
+available on that machine; TensorStudio won 42 local cases and lost 56. The
 strongest local wins were the simple NumPy convolution/pooling reference loops
 and some small JAX-dispatch-heavy eager cases. NumPy and JAX were faster for
 many elementwise, reduction, matrix multiplication, larger activation, and
@@ -426,15 +461,15 @@ Snapshot from that local run:
 
 | operation | shape | TensorStudio | NumPy | JAX CPU dispatch | TS vs NumPy | TS vs JAX |
 |---|---:|---:|---:|---:|---:|---:|
-| `sigmoid` | `(32,)` | 0.0154 ms | 0.0048 ms | 0.0855 ms | 0.3098x | 5.5404x |
-| `mean` | `(32,)` | 0.0187 ms | 0.0104 ms | 0.0143 ms | 0.5575x | 0.7616x |
-| `sum_axis1` | `(16, 16)` | 0.0172 ms | 0.0027 ms | 0.0136 ms | 0.1551x | 0.7942x |
-| `chain_relu` | `(128,)` | 0.0957 ms | 0.0057 ms | 0.1030 ms | 0.0596x | 1.0768x |
-| `matmul` | `(256, 256)` | 2.3777 ms | 0.4328 ms | 0.2489 ms | 0.1820x | 0.1047x |
-| `conv2d_3x3_padding1` | `(1, 1, 8, 8)` | 0.2216 ms | 1.3908 ms | 0.1073 ms | 6.2770x | 0.4841x |
-| `max_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0280 ms | 0.1846 ms | n/a | 6.5816x | n/a |
-| `avg_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0281 ms | 0.5595 ms | n/a | 19.8874x | n/a |
-| `elementwise_backward` | `(1024,)` | 2.8255 ms | n/a | n/a | n/a | n/a |
+| `sigmoid` | `(32,)` | 0.0151 ms | 0.0085 ms | 0.0788 ms | 0.5609x | 5.2163x |
+| `mean` | `(32,)` | 0.0162 ms | 0.0085 ms | 0.0160 ms | 0.5228x | 0.9868x |
+| `sum_axis1` | `(16, 16)` | 0.0162 ms | 0.0027 ms | 0.0148 ms | 0.1645x | 0.9159x |
+| `chain_relu` | `(128,)` | 0.0908 ms | 0.0063 ms | 0.0964 ms | 0.0699x | 1.0613x |
+| `matmul` | `(256, 256)` | 2.5442 ms | 0.4174 ms | 0.2867 ms | 0.1641x | 0.1127x |
+| `conv2d_3x3_padding1` | `(1, 1, 8, 8)` | 0.2048 ms | 1.4268 ms | 0.1191 ms | 6.9654x | 0.5814x |
+| `max_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0278 ms | 0.1944 ms | n/a | 6.9988x | n/a |
+| `avg_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0286 ms | 0.6478 ms | n/a | 22.6805x | n/a |
+| `elementwise_backward` | `(1024,)` | 3.0411 ms | n/a | n/a | n/a | n/a |
 
 Speedup is `competitor median / TensorStudio median`, so values above `1.0x`
 favor TensorStudio.

{tensorstudio-1.9.0 → tensorstudio-1.11.0}/README.md

@@ -7,15 +7,18 @@
 TensorStudio is a compact C++ tensor and autograd engine with a Python API for
 learning, experimentation, and lightweight ML workloads.
 
-TensorStudio `1.9.0` is a CPU-only stable API foundation with native C++
+TensorStudio `1.11.0` is a CPU-only stable API foundation with native C++
 threading, storage reuse, SIMD-friendly typed kernels, and optional
 CBLAS/Accelerate matrix multiplication when available. It adds native stable
 softmax/logsumexp, batched matrix multiplication, statistical reductions,
 boolean reductions, seeded random distributions, and a hardened eager autograd
 lifecycle. The neural-network layer now includes grouped/depthwise/1D/
 transposed convolution, normalization layers, embeddings, richer activations,
-initializers, additional losses, and model summaries. It is eager-only,
-intentionally compact, and not a replacement for mature ML frameworks.
+initializers, additional losses, and model summaries. The project layer adds
+dataset factories, deterministic train/validation splitting, metrics, callbacks,
+multi-format configs, checkpoint resume helpers, and starter project templates.
+It is eager-only, intentionally compact, and not a replacement for mature ML
+frameworks.
 
 ## Install
 
@@ -225,7 +228,7 @@ print(model.state_dict().keys())
 print(model.parameter_count())
 ```
 
-The `1.9.0` neural-network surface also includes initialization helpers,
+The neural-network surface also includes initialization helpers,
 normalization layers, embeddings, grouped/depthwise/1D/transposed convolution,
 adaptive/global pooling, richer activations, and model summaries:
 
@@ -249,6 +252,11 @@ classification workflows: Pillow-backed image IO, transform pipelines,
 deterministic augmentations, `ImageFolder` datasets, metrics, image grids,
 bounding-box drawing, and compact CNN classifiers running through native
 Conv2d/pooling kernels.
+The `1.11.0` vision surface adds batch-aware transforms, color jitter, random
+resized crop, rotation, affine transforms, cutout, mixup, CutMix, detection
+helpers, segmentation mask helpers, detection/segmentation folder datasets,
+ResNet-style blocks, MobileNet-style depthwise blocks, a compact UNet, and
+prediction/mask/feature-map visualization helpers.
 
 ```python
 import numpy as np
@@ -276,17 +284,22 @@ optimizer.step()
 print(ts.vision.accuracy(model(x), target))
 ```
 
-## DataLoader
+## Data And Metrics
 
 ```python
 import tensorstudio as ts
-from tensorstudio.data import DataLoader, TensorDataset
+from tensorstudio.data import DataLoader, from_arrays, train_val_split
 
-dataset = TensorDataset(ts.arange(6).reshape((6, 1)), ts.arange(6))
-loader = DataLoader(dataset, batch_size=2, shuffle=True, seed=42)
+dataset = from_arrays([[0.0], [1.0], [2.0], [3.0]], [[1.0], [3.0], [5.0], [7.0]])
+train_data, val_data = train_val_split(dataset, val_fraction=0.25, seed=42)
+loader = DataLoader(train_data, batch_size=2, shuffle=True, seed=42)
 
 for features, targets in loader:
     print(features.shape, targets.shape)
+
+prediction = ts.tensor([[0.2, 0.8], [0.7, 0.3]])
+target = ts.tensor([1, 0], dtype="int64")
+print(ts.metrics.accuracy(prediction, target))
 ```
 
 The v1 DataLoader is intentionally single-process so it works cleanly on
@@ -294,24 +307,44 @@ Windows without multiprocessing setup.
 
 ## Projects And Training
 
-`tensorstudio.project` provides project folders, JSON config, reusable trainers,
-safe NPZ weight files, and trusted full checkpoints:
+`tensorstudio.project` provides project folders, JSON/TOML/YAML config loading,
+deterministic seeding, reusable trainers, validation loops, callbacks, safe NPZ
+weight files, trusted full checkpoints, and generated starter templates:
 
 ```python
 import tensorstudio as ts
 from tensorstudio import nn, optim
-from tensorstudio.data import DataLoader, TensorDataset
-from tensorstudio.project import Project, ProjectConfig, Trainer, save_state_dict
+from tensorstudio.data import DataLoader, from_arrays, train_val_split
+from tensorstudio.project import (
+    CSVLogger,
+    CheckpointCallback,
+    LrLogger,
+    Project,
+    ProjectConfig,
+    Trainer,
+    save_state_dict,
+    seed_everything,
+)
 
-x = ts.tensor([[0.0], [1.0], [2.0], [3.0]])
-y = ts.tensor([[1.0], [3.0], [5.0], [7.0]])
+seed_everything(7)
+dataset = from_arrays([[0.0], [1.0], [2.0], [3.0]], [[1.0], [3.0], [5.0], [7.0]])
+train_data, val_data = train_val_split(dataset, val_fraction=0.25, seed=7)
 
 model = nn.Linear(1, 1)
-loader = DataLoader(TensorDataset(x, y), batch_size=2)
-trainer = Trainer(model, optim.SGD(model.parameters(), lr=0.05), nn.MSELoss())
+optimizer = optim.SGD(model.parameters(), lr=0.05)
+trainer = Trainer(model, optimizer, nn.MSELoss(), metric_fn=ts.metrics.mean_squared_error)
 project = Project("runs/linear", ProjectConfig(name="linear-regression", seed=7))
 
-history = trainer.fit(loader, epochs=50)
+history = trainer.fit(
+    DataLoader(train_data, batch_size=2),
+    epochs=50,
+    validation_loader=DataLoader(val_data, batch_size=1),
+    callbacks=[
+        LrLogger(),
+        CSVLogger(project.logs_dir / "history.csv"),
+        CheckpointCallback(project.checkpoints_dir / "best.tsmodel", save_best_only=True),
+    ],
+)
 save_state_dict(model, project.checkpoint_path("weights"))
 print(history.last)
 ```
@@ -346,11 +379,11 @@ Run the loose local regression thresholds with:
 python benchmark_all.py --check-thresholds
 ```
 
-On one Windows CPython 3.10 development run reporting `1.9.0`, with
+On one Windows CPython 3.10 development run reporting `1.11.0`, with
 TensorStudio threads enabled, storage pooling enabled, SSE2 autovectorization
 reported, and no BLAS provider found, TensorStudio beat NumPy on 7 local
 benchmark cases and lost on 96 NumPy-comparable cases. JAX CPU dispatch was
-available on that machine; TensorStudio won 55 local cases and lost 43. The
+available on that machine; TensorStudio won 42 local cases and lost 56. The
 strongest local wins were the simple NumPy convolution/pooling reference loops
 and some small JAX-dispatch-heavy eager cases. NumPy and JAX were faster for
 many elementwise, reduction, matrix multiplication, larger activation, and
@@ -362,15 +395,15 @@ Snapshot from that local run:
 
 | operation | shape | TensorStudio | NumPy | JAX CPU dispatch | TS vs NumPy | TS vs JAX |
 |---|---:|---:|---:|---:|---:|---:|
-| `sigmoid` | `(32,)` | 0.0154 ms | 0.0048 ms | 0.0855 ms | 0.3098x | 5.5404x |
-| `mean` | `(32,)` | 0.0187 ms | 0.0104 ms | 0.0143 ms | 0.5575x | 0.7616x |
-| `sum_axis1` | `(16, 16)` | 0.0172 ms | 0.0027 ms | 0.0136 ms | 0.1551x | 0.7942x |
-| `chain_relu` | `(128,)` | 0.0957 ms | 0.0057 ms | 0.1030 ms | 0.0596x | 1.0768x |
-| `matmul` | `(256, 256)` | 2.3777 ms | 0.4328 ms | 0.2489 ms | 0.1820x | 0.1047x |
-| `conv2d_3x3_padding1` | `(1, 1, 8, 8)` | 0.2216 ms | 1.3908 ms | 0.1073 ms | 6.2770x | 0.4841x |
-| `max_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0280 ms | 0.1846 ms | n/a | 6.5816x | n/a |
-| `avg_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0281 ms | 0.5595 ms | n/a | 19.8874x | n/a |
-| `elementwise_backward` | `(1024,)` | 2.8255 ms | n/a | n/a | n/a | n/a |
+| `sigmoid` | `(32,)` | 0.0151 ms | 0.0085 ms | 0.0788 ms | 0.5609x | 5.2163x |
+| `mean` | `(32,)` | 0.0162 ms | 0.0085 ms | 0.0160 ms | 0.5228x | 0.9868x |
+| `sum_axis1` | `(16, 16)` | 0.0162 ms | 0.0027 ms | 0.0148 ms | 0.1645x | 0.9159x |
+| `chain_relu` | `(128,)` | 0.0908 ms | 0.0063 ms | 0.0964 ms | 0.0699x | 1.0613x |
+| `matmul` | `(256, 256)` | 2.5442 ms | 0.4174 ms | 0.2867 ms | 0.1641x | 0.1127x |
+| `conv2d_3x3_padding1` | `(1, 1, 8, 8)` | 0.2048 ms | 1.4268 ms | 0.1191 ms | 6.9654x | 0.5814x |
+| `max_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0278 ms | 0.1944 ms | n/a | 6.9988x | n/a |
+| `avg_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0286 ms | 0.6478 ms | n/a | 22.6805x | n/a |
+| `elementwise_backward` | `(1024,)` | 3.0411 ms | n/a | n/a | n/a | n/a |
 
 Speedup is `competitor median / TensorStudio median`, so values above `1.0x`
 favor TensorStudio.