Merge pull request #154 from google/improve-reference

Improve reference

.gitignore

@@ -6,6 +6,7 @@ tmp
 build_package
 .ipynb_checkpoints
 tmp_*
+.cache/
 .env
 
 # benchmark outputs
@@ -18,7 +19,7 @@ profile.*
 site/
 
 # Build outputs
-build
+build/
 dist
 setup.py
 temporian.egg-info

.vscode/settings.json

@@ -17,12 +17,17 @@
   "[python]": {
     "editor.defaultFormatter": "ms-python.black-formatter"
   },
+  "black-formatter.path": [
+    "${workspaceFolder}/.venv/bin/black"
+  ],
   "editor.tabSize": 4,
   "editor.formatOnSave": true,
   "python.linting.flake8Enabled": false,
   "python.linting.pylintEnabled": true,
   "python.linting.enabled": true,
-  "editor.rulers": [80],
+  "editor.rulers": [
+    80
+  ],
   "editor.codeActionsOnSave": {
     "source.organizeImports": false
   },
@@ -82,4 +87,4 @@
     "span": "cpp",
     "algorithm": "cpp"
   }
-}
+}

README.md

@@ -43,22 +43,22 @@ date,feature_1,feature_2
 2023-02-01,30.0,5.0
 ```
 
-Check the [Getting Started tutorial](https://temporian.readthedocs.io/en/latest/tutorials/getting_started/) to try it out!
+Check the [Getting Started tutorial](https://temporian.readthedocs.io/en/stable/tutorials/getting_started/) to try it out!
 
 ## Key features
 
 These are what set Temporian apart.
 
 - **Simple and powerful API**: Temporian exports high level operations making processing complex programs short and ready to read.
 - **Flexible data model**: Temporian models temporal data as a sequence of events, supporting non-uniform sampling timestamps seamlessly.
-- **Prevents modeling errors**: Temporian programs are guaranteed not to have future leakage unless the user calls the `leak` function, ensuring that models are not trained on future data.
+- **Prevents modeling errors**: Temporian programs are guaranteed not to have future leakage unless explicitly specified, ensuring that models are not trained on future data.
 - **Iterative development**: Temporian can be used to develop preprocessing pipelines in Colab or local notebooks, allowing users to visualize results each step of the way to identify and correct errors early on.
 - **Efficient and well-tested implementations**: Temporian contains efficient and well-tested implementations of a variety of temporal data processing functions. For instance, our implementation of window operators is **x2000** faster than the same function implemented with NumPy.
-- **Wide range of preprocessing functions**: Temporian contains a wide range of preprocessing functions, including moving window operations, lagging, calendar features, arithmetic operations, index manipulation and propagation, resampling, and more. For a full list of the available operators, see the [operators documentation](https://temporian.readthedocs.io/en/latest/reference/temporian/core/operators/all_operators/).
+- **Wide range of preprocessing functions**: Temporian contains a wide range of preprocessing functions, including moving window operations, lagging, calendar features, arithmetic operations, index manipulation and propagation, resampling, and more. For a full list of the available operators, see the [operators documentation](https://temporian.readthedocs.io/en/stable/reference/).
 
 ## Documentation
 
-The official documentation is available at [temporian.readthedocs.io](https://temporian.readthedocs.io/en/latest/).
+The official documentation is available at [temporian.readthedocs.io](https://temporian.readthedocs.io/en/stable/).
 
 ## Contributing
 

docs/gen_ref_pages.py

@@ -5,85 +5,107 @@
 """
 
 from pathlib import Path
+from typing import Set, Tuple
 
 import mkdocs_gen_files
 
 nav = mkdocs_gen_files.Nav()
 
 SRC_PATH = Path("temporian")
 
-paths = set()
+# Stores symbol and path of each public API member
+members: Set[Tuple[str, Path]] = set()
+
 non_parsable_imports = []
 
-with open("temporian/__init__.py", "r", encoding="utf8") as f:
-    lines = f.read().splitlines()
+# We need to be able to parse other files to allow wildcard imports
+# Storing pair of (prefix, path) to parse in a stack
+files_to_parse = [(None, SRC_PATH / "__init__.py")]
+
+while files_to_parse:
+    prefix, file = files_to_parse.pop()
+
+    with open(file, "r", encoding="utf8") as f:
+        lines = f.read().splitlines()
+
+    for line in lines:
+        words = line.split(" ")
+
+        # It is an import statement
+        if words[0] == "from":
+            # Remove trailing "as <name>" if it exists and save symbol's name
+            symbol = None
+            if words[-2] == "as":
+                # If symbol was renamed to a private name, skip it
+                if words[-1].startswith("_"):
+                    continue
 
-for line in lines:
-    words = line.split(" ")
+                symbol = words[-1]
+                words = words[:-2]
 
-    # It is an import statement
-    if words[0] == "from":
-        # Remove trailing "as <name>" if it exists
-        if words[-2] == "as":
-            # If symbol was renamed to a private name, skip it
-            if words[-1].startswith("_"):
-                continue
+            # `words` is now in the form "from module.submodule import symbol"
+            if words[-2] == "import":
+                name = words[-1]
 
-            words = words[:-2]
+                # We only allow wildcard imports from modules explicitly named
+                # api_symbols to prevent unwanted names in the public API
+                if name == "*":
+                    module_path = Path(words[1].replace(".", "/")).with_suffix(
+                        ".py"
+                    )
+                    if module_path.stem == "api_symbols":
+                        new_prefix = (
+                            (prefix + ".") if prefix else ""
+                        ) + module_path.parent.name
+                        files_to_parse.append((new_prefix, module_path))
+                        continue
 
-        # It is a single-symbol import like "from <module> import <symbol>"
-        if words[-2] == "import":
-            module_path = Path(words[1].replace(".", "/"))
+                    non_parsable_imports.append(line)
+                    continue
 
-            # Check if the import is a dir module
-            module_path_with_suffix = module_path / words[-1]
-            if module_path_with_suffix.exists():
-                module_path = module_path_with_suffix
+                # If symbol wasn't renamed, use its imported name
+                if symbol is None:
+                    symbol = name
 
-            # Check if the import is a file module
-            module_path_with_suffix = module_path / (words[-1] + ".py")
-            if module_path_with_suffix.exists():
-                module_path = module_path_with_suffix.with_suffix("")
+                path = Path(words[1].replace(".", "/")) / name
 
-            # If it's not a module import it is a normal symbol import
-            # (function, class, etc.) so we add its whole module to the docs
+                if prefix:
+                    symbol = prefix + "." + symbol
 
-            paths.add(module_path)
+                members.add((symbol, path))
 
-        else:
-            non_parsable_imports.append(line)
+            # It is a multi-symbol import statement, error will be raised below
+            else:
+                non_parsable_imports.append(line)
 
 if non_parsable_imports:
     raise RuntimeError(
         "`gen_ref_pages` failed to parse the following import statements in"
         f" the top-level __init__.py file: {non_parsable_imports}. Import"
         " statements in the top-level module must import a single symbol each,"
-        " in the form `from <module> import <symbol>` or `from <module> import"
-        " <symbol> as <name>`."
+        " in the form `from <module> import <symbol>`, `from <module> import"
+        " <symbol> as <name>`, or `from <module> import *`."
     )
 
-for path in sorted(paths):
-    if path.parent.name not in ["test", "tests"]:
-        module_path = path.relative_to(SRC_PATH.parent).with_suffix("")
-        doc_path = path.relative_to(SRC_PATH.parent).with_suffix(".md")
-        full_doc_path = Path("reference", doc_path)
+nav["temporian"] = "index.md"
 
-        parts = list(module_path.parts)
+for symbol, path in sorted(members):
+    symbol_path = Path(symbol.replace(".", "/"))
+    symbol_name = symbol_path.name
+    src_path = SRC_PATH / symbol_name
 
-        if parts[-1] == "__init__":
-            parts = parts[:-1]
-            doc_path = doc_path.with_name("index.md")
-            full_doc_path = full_doc_path.with_name("index.md")
-        elif parts[-1] == "__main__":
-            continue
+    doc_path = SRC_PATH / symbol_path
+    parts = list(doc_path.parts)
+    doc_path = doc_path.with_suffix(".md")
+    full_doc_path = Path("reference", doc_path)
 
-        nav[parts] = doc_path.as_posix()
+    nav[parts] = doc_path.as_posix()
 
-        with mkdocs_gen_files.open(full_doc_path, "w") as fd:
-            identifier = ".".join(parts)
-            print("::: " + identifier, file=fd)
+    with mkdocs_gen_files.open(full_doc_path, "w") as fd:
+        identifier = ".".join(list(src_path.parts))
+        print("::: " + identifier, file=fd)
 
-        mkdocs_gen_files.set_edit_path(full_doc_path, path)
+    mkdocs_gen_files.set_edit_path(full_doc_path, path)
 
-with mkdocs_gen_files.open("reference/index.md", "w") as nav_file:
+with mkdocs_gen_files.open("reference/SUMMARY.md", "w") as nav_file:
     nav_file.writelines(nav.build_literate_nav())

docs/mkdocs.yml

@@ -1,5 +1,9 @@
 site_name: Temporian
+site_url: https://temporian.readthedocs.io/en/stable/
+site_description: A Python package for feature engineering of temporal data.
+
 repo_url: https://github.com/google/temporian
+edit_uri_template: tree/main/temporian/{path_noext}
 
 # Theme
 theme:
@@ -48,14 +52,8 @@ nav:
 plugins:
   - search
   - exclude-search:
-      # Avoid this since it excludes all ipynb
-      exclude_unreferenced: false
-      # Exclude reference's Index page
       exclude:
-        - reference
-      # Include all pages inside reference
-      ignore:
-        - reference/*
+        - reference/SUMMARY
   - autorefs
   - gen-files:
       scripts:
@@ -65,6 +63,7 @@ plugins:
         - gen_ref_pages.py
   - literate-nav:
       nav_file: SUMMARY.md
+  - social
   - mkdocs-jupyter:
       # Execute notebooks when building docs (set to true when temporian runs in a notebook w/o the start_notebook script).
       execute: false
@@ -76,24 +75,28 @@ plugins:
       default_handler: python
       handlers:
         python:
-          paths: [temporian]
+          paths: [..]
           import:
             - https://docs.python.org/3/objects.inv
           options:
             # https://mkdocstrings.github.io/python/usage/#globallocal-options
             docstring_style: google
             heading_level: 1
             members_order: source
-            show_source: true
-            show_root_heading: true
-            show_category_heading: true
-            show_submodules: true
+            show_source: false
+            show_submodules: false
             merge_init_into_class: false
             show_signature: true
             separate_signature: true
-            show_signature_annotations: false
-            show_if_no_docstring: false # TODO: TBD if we want this enabled
-            group_by_category: false
+            show_signature_annotations: true
+            show_if_no_docstring: false
+            group_by_category: true
+            show_category_heading: false
+            show_root_heading: true
+            # show_root_toc_entry: false
+            # show_symbol_type_heading: false
+            # preload_modules: [temporian.core.operators]
+            # allow_inspection: true
 
 # Customization for Markdown
 markdown_extensions:

docs/requirements.txt

@@ -8,4 +8,6 @@ mkdocs-section-index==0.3.5
 mkdocs-jupyter===0.24.1
 mkdocs-exclude-search===0.6.5
 black==22.12.0
-griffe==0.26.0
+griffe==0.26.0
+cairosvg==2.7.0
+pillow==9.5.0

docs/src/3_minutes.md

@@ -2,15 +2,15 @@
 
 This is a _very_ quick introduction to how Temporian works. For a complete tour of its capabilities, please refer to the [User Guide](../user_guide).
 
-## Events and `EventSets`
+## Events and [`EventSets`][temporian.EventSet]
 
 The most basic unit of data in Temporian is an **event**. An event consists of a timestamp and a set of feature values.
 
-Events are not handled individually. Instead, events are grouped together into an **[EventSet](../reference/temporian/implementation/numpy/data/event_set)**.
+Events are not handled individually. Instead, events are grouped together into an **[EventSet][temporian.EventSet]**.
 
-`EventSet`s are the main data structure in Temporian, and represent **[multivariate time sequences](../user_guide/#what-is-temporal-data)**. Note that "multivariate" indicates that each event in the time sequence holds several feature values, and "sequence" indicates that the events are not necessarily sampled at a uniform rate (in which case we would call it a time "series").
+[`EventSets`][temporian.EventSet] are the main data structure in Temporian, and represent **[multivariate time sequences](../user_guide/#what-is-temporal-data)**. Note that "multivariate" indicates that each event in the time sequence holds several feature values, and "sequence" indicates that the events are not necessarily sampled at a uniform rate (in which case we would call it a time "series").
 
-You can create an `EventSet` from a pandas DataFrame, NumPy arrays, CSV files, and more. Here is an example of an `EventSet` containing four events and three features:
+You can create an [`EventSet`][temporian.EventSet] from a pandas DataFrame, NumPy arrays, CSV files, and more. Here is an example of an [`EventSet`][temporian.EventSet] containing four events and three features:
 
 ```python
 >>> evset = tp.event_set(
@@ -25,21 +25,21 @@ You can create an `EventSet` from a pandas DataFrame, NumPy arrays, CSV files, a
 
 ```
 
-An `EventSet` can hold one or several time sequences, depending on what its **[index](../user_guide/#index-horizontal-and-vertical-operators)** is.
+An [`EventSet`][temporian.EventSet] can hold one or several time sequences, depending on what its **[index](../user_guide/#index-horizontal-and-vertical-operators)** is.
 
-If the `EventSet` has no index, it will hold a single time sequence, which means that all events will be considered part of the same group and will interact with each other when operators are applied to the `EventSet`.
+If the [`EventSet`][temporian.EventSet] has no index, it will hold a single time sequence, which means that all events will be considered part of the same group and will interact with each other when operators are applied to the [`EventSet`][temporian.EventSet].
 
-If the `EventSet` has one (or many) indexes, it will hold one time sequence for each unique value (or unique combination of values) of the indexes, the events will be grouped by their index value, and operators applied to the `EventSet` will be applied to each time sequence independently.
+If the [`EventSet`][temporian.EventSet] has one (or many) indexes, it will hold one time sequence for each unique value (or unique combination of values) of the indexes, the events will be grouped by their index value, and operators applied to the [`EventSet`][temporian.EventSet] will be applied to each time sequence independently.
 
-## Graph, `Nodes` and Operators
+## Graph, [`Nodes`][temporian.Node] and Operators
 
 There are two big phases in any Temporian script: graph **definition** and **evaluation**. This is a common pattern in computing libraries, and it allows us to perform optimizations before the graph is evaluated, share Temporian programs across different platforms, and more.
 
-A graph is created by using **operators**. For example, the [`tp.simple_moving_average()`](../reference/temporian/core/operators/window/simple_moving_average) operator computes the [simple moving average](https://en.wikipedia.org/wiki/Moving_average) of each feature in an `EventSet`. You can find documentation for all available operators [here](../reference/temporian/core/operators/all_operators).
+A graph is created by using **operators**. For example, the [`tp.simple_moving_average()`][temporian.simple_moving_average] operator computes the [simple moving average](https://en.wikipedia.org/wiki/Moving_average) of each feature in an [`EventSet`][temporian.EventSet]. You can find documentation for all available operators [here](../reference/).
 
 Note that when calling operators you are only defining the graph - i.e., you are telling Temporian what operations you want to perform on your data, but those operations are not yet being performed.
 
-Operators are not applied directly to `EventSet`s, but to **[Nodes](../reference/temporian/core/data/node)**. You can think of a `Node` as the placeholder for an `EventSet` in the graph. When applying operators to `Node`s, you get back new `Node`s that are placeholders for the results of those operations. You can create arbitrarily complex graphs by combining operators and nodes.
+Operators are not applied directly to [`EventSets`][temporian.EventSet], but to **[Nodes][temporian.Node]**. You can think of a [`Node`][temporian.Node] as the placeholder for an [`EventSet`][temporian.EventSet] in the graph. When applying operators to [`Nodes`][temporian.Node], you get back new [`Nodes`][temporian.Node] that are placeholders for the results of those operations. You can create arbitrarily complex graphs by combining operators and nodes.
 
 ```python
 >>> # Obtain the Node corresponding to the EventSet we created above
@@ -53,14 +53,14 @@ Operators are not applied directly to `EventSet`s, but to **[Nodes](../reference
 
 <!-- TODO: add image of the generated graph -->
 
-Your graph can now be run by calling [`evaluate()`](../reference/temporian/core/data/node/#temporian.core.data.node.Node.evaluate) on any `Node` in the graph, which will perform all necessary operations and return the resulting `EventSet`.
+Your graph can now be run by calling [`.evaluate()`][temporian.Node.evaluate] on any [`Node`][temporian.Node] in the graph, which will perform all necessary operations and return the resulting [`EventSet`][temporian.EventSet].
 
 ```python
 >>> result = addition_lagged.evaluate(evset)
 
 ```
 
-Note that you need to pass the `EventSet`s that correspond to the source `Node`s in the graph to `evaluate()` (since those are not part of the graph definition). Also, several `Node`s can be evaluated at the same time by calling [`tp.evaluate()`](../reference/temporian/core/evaluation/#temporian.core.evaluation.evaluate) directly.
+Note that you need to pass the [`EventSets`][temporian.EventSet] that correspond to the source [`Nodes`][temporian.Node] in the graph to [`.evaluate()`][temporian.Node.evaluate] (since those are not part of the graph definition). Also, several [`Nodes`][temporian.Node] can be evaluated at the same time by calling [`tp.evaluate()`][temporian.evaluate] directly.
 
 🥳 Congratulations! You're all set to write your first pieces of Temporian code.
 

docs/src/css/custom.css

@@ -1,4 +1,12 @@
 [data-md-color-scheme="temporian"] {
     --md-primary-fg-color--light: #24201D;
     --md-primary-fg-color--dark: #24201D;
+    background-color: red;
+    --md-default-bg-color: red;
 }
+
+/* Change background color when dark mode */
+[data-md-color-scheme="slate"] {
+    --md-default-bg-color: #121212;
+    --md-code-bg-color: #212121;
+}

docs/src/reference/index.md

@@ -0,0 +1,5 @@
+# API Reference
+
+Temporian's API Reference.
+
+<!-- TODO: write reference home page. Include table of operators. -->