2. Graph objects contain descriptions of each valid property as Python docstrings, with a [full API reference available](https://plotly.com/python-api-reference/). You can use these docstrings in the development environment of your choice to learn about the available properties as an alternative to consulting the online [Full Reference](https://plotly.com/python/reference/).

The figures created, manipulated and rendered by the plotly Python library are [represented by tree-like data structures](https://plotly.com/python/figure-structure/) which are automatically serialized to JSON for rendering by the Plotly.js JavaScript library. These trees are composed of named nodes called "attributes", with their structure defined by the Plotly.js figure schema, which is available in [machine-readable form](https://raw.githubusercontent.com/plotly/plotly.js/master/dist/plot-schema.json). **The `plotly.graph_objects` module (typically imported as `go`) contains an [automatically-generated hierarchy of Python classes](https://plotly.com/python-api-reference/plotly.graph_objects.html#graph-objects) which represent non-leaf nodes in this figure schema. The term "graph objects" refers to instances of these classes. **

The primary classes defined in the `plotly.graph_objects` module are [`Figure`](https://plotly.com/python-api-reference/generated/plotly.graph_objects.Figure.html) and an [`ipywidgets`-compatible variant called `FigureWidget`](https://plotly.com/python/figurewidget/), which both represent entire figures. Instances of these classes have many convenience methods for Pythonically [manipulating their attributes](https://plotly.com/python/creating-and-updating-figures/) (e.g. `.update_layout()` or `.add_trace()`, which all accept ["magic underscore" notation](https://plotly.com/python/creating-and-updating-figures/#magic-underscore-notation)) as well as [rendering them](https://plotly.com/python/renderers/) (e.g. `.show()`) and [exporting them to various formats](https://plotly.com/python/static-image-export/) (e.g. `.to_json()` or `.write_image()` or `.write_html()`).

Every non-leaf attribute of a figure is represented by an instance of a class in the `plotly.graph_objects` hierarchy. For example, a figure `fig` can have an attribute `layout.margin`, which contains attributes `t`, `l`, `b` and `r` which are leaves of the tree: they have no children. The field at `fig.layout` is an object of class [`plotly.graph_objects.Layout`](https://plotly.com/python-api-reference/generated/plotly.graph_objects.Layout.html) and `fig.layout.margin` is an object of class `plotly.graph_objects.layout.Margin` which represents the `margin` node, and it has fields `t`, `l`, `b` and `r`, containing the values of the respective leaf-nodes. Note that specifying all of these values can be done without creating intermediate objects using ["magic underscore" notation](https://plotly.com/python/creating-and-updating-figures/#magic-underscore-notation): `go.Figure(layout_margin=dict(t=10, b=10, r=10, l=10))`.

The objects contained in the list which is the [value of the attribute `data` are called "traces"](https://plotly.com/python/figure-structure/), and can be of one of more than 40 possible types, each of which has a corresponding class in `plotly.graph_objects`. For example, traces of type `scatter` are represented by instances of the class `plotly.graph_objects.Scatter`. This means that a figure constructed as `go.Figure(data=[go.Scatter(x=[1,2], y=[3,4)])` will have the JSON representation `{"data": [{"type": "scatter", "x": [1,2], "y": [3,4]}]}`.

2. Graph objects contain descriptions of each valid property as Python docstrings, with a [full API reference available](https://plotly.com/python-api-reference/). You can use these docstrings in the development environment of your choice to learn about the available properties as an alternative to consulting the online [Full Reference](https://plotly.com/python/reference/).

The recommended way to create figures is using the [functions in the plotly.express module](https://plotly.com/python-api-reference/), [collectively known as Plotly Express](https://plotly.com/python/plotly-express/), which all return instances of `plotly.graph_objects.Figure`, so every figure produced with the plotly library, actually uses graph objects under the hood, unless manually constructed out of dictionaries.

That said, certain kinds of figures are not yet possible to create with Plotly Express, such as figures that use certain 3D trace-types like [`mesh`](https://plotly.com/python/3d-mesh/) or [`isosurface`](https://plotly.com/python/3d-isosurface-plots/). In addition, certain figures are cumbersome to create by starting from a figure created with Plotly Express, for example figures with [subplots of different types](https://plotly.com/python/mixed-subplots/), [dual-axis plots](https://plotly.com/python/multiple-axes/), or [faceted plots](https://plotly.com/python/facet-plots/) with multiple different types of traces. To construct such figures, it can be easier to start from an empty `plotly.graph_objects.Figure` object (or one configured with subplots via the [make_subplots() function](https://plotly.com/python/subplots/)) and progressively add traces and update attributes as above.

Note that the figures produced by Plotly Express **in a single function-call** are [easy to customize at creation-time](https://plotly.com/python/styling-plotly-express/), and to [manipulate after creation](https://plotly.com/python/creating-and-updating-figures/) using the `update_*` and `add_*` methods. The figures produced by Plotly Express can always be built from the ground up using graph objects, but this approach typically takes **5-100 lines of code rather than 1**. Here is a simple example of how to produce the same figure object from the same data, once with Plotly Express and once without. The data in this example is in "long form" but [Plotly Express also accepts data in "wide form"](https://plotly.com/python/wide-form/) and the line-count savings from Plotly Express over graph objects are comparable. More complex figures such as [parallel coordinates](https://plotly.com/python/parallel-coordinates-plot/), [facet plots](https://plotly.com/python/facet-plots/) or [animations](https://plotly.com/python/animations/) require many more lines of figure-specific graph objects code, whereas switching from one representation to another with Plotly Express usually involves changing just a few characters.