Merge pull request #1 from JonasLoos/custom-plotting

Custom plotting

Author: JonasLoos

.gitignore

@@ -26,6 +26,7 @@ share/python-wheels/
 .installed.cfg
 *.egg
 MANIFEST
+uv.lock
 
 # PyInstaller
 #  Usually these files are written by a python script from a template

README.md

@@ -1,91 +1,50 @@
 # trainplot
 
-Dynamically updating plots in Jupyter notebooks, e.g. for visualizing training progress. Inspired by [livelossplot](https://github.com/stared/livelossplot), and aims to be easier to use with better jupyter notebook support.
-
-
-## Installation
+Dynamically updating plots in Jupyter notebooks, e.g. for visualizing machine learning training progress.
 
 ```bash
+
 pip install trainplot
 ```
 
+<p align="center"><img src="https://github.com/user-attachments/assets/03bd661c-37d7-41a4-ba91-891f57ebfcf8" width="400" center></p>
+
+
 
 ## Usage
 
-This is a simple example ([example notebook](examples/basic-example.ipynb)):
+Basic usage:
 
 ```python
 from trainplot import plot
-from time import sleep
 
-for i in range(50):
-    plot(loss = 1/(i+1), acc = 1-1/(.01*i**2+1))
-    sleep(.1)
+for i in range(100):
+    loss = ...
+    acc = ...
+    plot(loss=loss, accuracy=acc)
 ```
 
-<img src="https://github.com/JonasLoos/trainplot/assets/33965649/935e8d52-0c37-4469-9cb8-24fa77b467ff" width="500">
-
----
-
-Example for the tf/keras callback ([example notebook](examples/tf-keras-mnist-example.ipynb)):
+If you use keras, you can use the `TrainPlotKerasCallback`:
 
 ```python
 from trainplot import TrainPlotKeras
 
 model = ...
-model.fit(x_train, y_train, validation_data=(x_test, y_test), epochs=10, callbacks=[TrainPlotKeras()])
-```
-
-<img src="https://github.com/JonasLoos/trainplot/assets/33965649/4ddff79a-978e-434c-a6c3-571cf48c0892" width="500">
-
----
-
-It also works together with e.g. `tqdm.notebook` and printing ([example notebook](examples/different-output-example.ipynb)):
-
-```python
-from trainplot import plot
-from tqdm.notebook import trange
-from time import sleep
-
-for i in trange(50):
-    plot(i=i, root=i**.5)
-    if i % 10 == 0:
-        print(f'currently at {i} iterations')
-    sleep(0.1)
+model.fit(x_train, y_train, validation_data=(x_test, y_test), epochs=10, callbacks=[TrainPlotKerasCallback()])
 ```
 
-<img src="https://github.com/JonasLoos/trainplot/assets/33965649/7571efab-7a3f-4414-b537-a2dffd9e1bec" width="400">
+For more examples, see the [`examples`](examples/) folder.
 
----
-
-You can make use of a TrainPlot object to add a bunch of custumizations ([example notebook](examples/4plots-example.ipynb)):
-
-```python
-from trainplot import TrainPlot
-from time import sleep
-
-tp = TrainPlot(
-    update_period=.2,
-    fig_args=dict(nrows=2, ncols=2, figsize=(10, 8), gridspec_kw={'height_ratios': [1, 1], 'width_ratios': [1, 1]}),
-    plot_pos={'loss': (0, 0, 0), 'accuracy': (0, 1, 0), 'val_loss': (1, 0, 0), 'val_accuracy': (1, 1, 0)},
-    plot_args={'loss': {'color': 'orange'}, 'accuracy': {'color': 'green'}, 'val_loss': {'color': 'orange', 'label': 'validation loss'}, 'val_accuracy': {'color': 'green', 'label': 'validation accuracy'}},
-)
-
-for i in range(100, 200):
-    tp(step=i, loss=(i/100-2)**4, accuracy=i/2, val_loss=(i/100-2.1)**4, val_accuracy=i/2.1)
-    sleep(0.1)
-```
 
-<img src="https://github.com/JonasLoos/trainplot/assets/33965649/599314e2-d1c1-4044-a915-6316722a2324" width="600">
+## Features
 
----
+* **Lightweight**: No external plotting dependencies
+* **Custom rendering**: Uses HTML5 Canvas for fast, smooth updates
+* **Multiple series**: Automatically handles multiple data series with different colors
+* **Real-time updates**: Configurable update periods to balance performance and responsiveness
+* **Keras support**: Built-in callback for TensorFlow/Keras models
 
-More:
-* When using a Trainplot object, you can also put the plot into a separate cell than the training loop: [example notebook](examples/separate-output-example.ipynb)
-* Experimental plotly support (`from trainplot.trainplot import TrainPlotPlotlyExperimental`): [example notebook](examples/plotly-example.ipynb)
 
 ## How it works
 
-Trainplot outputs the matplotlib figure to an `ipywidgets.Output` widget, so it doesn't interfere with other outputs like `tqdm` or print statements. To avoid wasting resources and flickering, the figure is only updated with a given `update_period`.
-A `post_run_cell` callback is added to the `IPython` instance, so that all updated TrainPlot figures include all new data when a cell execution is finished.
-When using `trainplot.plot`, a TrainPlot object is created for the current cell and cell-execution-count.
+Trainplot uses a custom HTML5 Canvas-based plotting solution that renders directly in Jupyter notebooks. For synchronization between Python and the JavaScript-based plotting function, [`anywidget`](https://github.com/manzt/anywidget) is used. To avoid wasting resources and flickering, the plot is only updated with a given `update_period`. A `post_run_cell` callback is added to the `IPython` instance, so that all updated TrainPlot figures include all new data when a cell execution is finished. When using `trainplot.plot`, a TrainPlot object is created for the current cell.

examples/4plots-example.ipynb

@@ -27,7 +27,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "You could do the same also by using a trainplot object:"
+    "You can do the same also by using a trainplot object:"
    ]
   },
   {
@@ -55,7 +55,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "base",
+   "display_name": ".venv",
    "language": "python",
    "name": "python3"
   },
@@ -69,7 +69,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.4"
+   "version": "3.13.3"
   }
  },
  "nbformat": 4,

examples/basic-example.ipynb

@@ -4,9 +4,9 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# Different Output Example\n",
+    "# Different Outputs Example\n",
     "\n",
-    "Here is an example, where tqdm and print statements write to the cell output at the same time as the plot is updated"
+    "Trainplot doesn't interfere with the output of other libraries or print statements."
    ]
   },
   {
@@ -21,15 +21,14 @@
     "\n",
     "for i in trange(50):\n",
     "    plot(i=i, root=i**.5)\n",
-    "    if i % 10 == 0:\n",
-    "        print(f'currently at {i} iterations')\n",
+    "    print(f'currently at {i} iterations', end='\\r')\n",
     "    sleep(0.1)"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "py311",
+   "display_name": ".venv",
    "language": "python",
    "name": "python3"
   },
@@ -43,7 +42,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.4"
+   "version": "3.12.3"
   }
  },
  "nbformat": 4,

examples/different-output-example.ipynb

@@ -6,12 +6,12 @@
    "source": [
     "# Separate Output Example\n",
     "\n",
-    "Here is a simple example, where the plotting output is in a separate cell from where the \"training loop\" is running."
+    "Ttrainplot supports multiple trainplot instances at the same time. They don't even have to be rendered and updated in the same cell."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -31,7 +31,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 3,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -44,7 +44,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "py311",
+   "display_name": ".venv",
    "language": "python",
    "name": "python3"
   },
@@ -58,7 +58,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.4"
+   "version": "3.12.3"
   }
  },
  "nbformat": 4,