[ENH] Add lazy imports to speed up the time taken to load pyjanitor (part 2) (#1180)

* plugged changes from commits from 3225b19 to 9d2732b(origin/feature/conditional-imports) to latest dev

* added lazy_imports.md and tuna image

* changes made in AUTHORS.md and CHANGELOG.md

* __version__ set be 0.23.1, lazy_loader is added to .requirements/base.in

Co-authored-by: Ubuntu <[email protected]>

Author: asmirnov69

.requirements/base.in

@@ -6,3 +6,4 @@ natsort
 pandas_flavor
 multipledispatch
 scipy
+lazy_loader

AUTHORS.md

@@ -109,3 +109,4 @@ Contributors
 - [@ethompsy](https://github.com/ethompsy) | [contributions](https://github.com/pyjanitor-devs/pyjanitor/issues?q=is%3Aclosed+mentions%3Aethompsy)
 - [@apatao](https://github.com/apatao) | [contributions](https://github.com/pyjanitor-devs/pyjanitor/issues?q=is%3Aclosed+mentions%3Aapatao)
 - [@OdinTech3](https://github.com/OdinTech3) | [contributions](https://github.com/pyjanitor-devs/pyjanitor/pull/1094)
+- [@asmirnov69](https://github.com/asmirnov69) | [contributions](https://github.com/pyjanitor-devs/pyjanitor/issues/1059)

CHANGELOG.md

@@ -2,6 +2,7 @@
 
 ## [Unreleased]
 
+-   [ENH] Add lazy imports to speed up the time taken to load pyjanitor (part 2)
 -   [DOC] Updated developer guide docs.
 -   [ENH] Allow column selection/renaming within conditional_join. Issue #1102. Also allow first or last match. Issue #1020 @samukweku.
 -   [ENH] New decorator `deprecated_kwargs` for breaking API. #1103 @Zeroto521

environment-dev.yml

@@ -55,3 +55,4 @@ dependencies:
   - xorg-libxrender
   - pip:
       - mknotebooks
+      - lazy-loader

janitor/__init__.py

@@ -1,15 +1,24 @@
-"""Top-level janitor API lives here."""
-try:
-    import janitor.xarray  # noqa: F401
-except ImportError:
-    pass
+"""Top-level janitor API lives here.
+
+Lazy loading used here to speed up imports.
+"""
+import lazy_loader as lazy
 
+
+from .accessors import *  # noqa: F403, F401
 from .functions import *  # noqa: F403, F401
 from .io import *  # noqa: F403, F401
 from .math import *  # noqa: F403, F401
 from .ml import get_features_targets as _get_features_targets
 from .utils import refactored_function
-from .accessors import *  # noqa: F403, F401
+
+
+_ = lazy.load("pandas_flavor")
+
+try:
+    jxr = lazy.load("janitor.xarray")  # noqa: F401
+except ImportError:
+    pass
 
 
 @refactored_function(

janitor/accessors/__init__.py

@@ -1,2 +1,17 @@
-"""Top-level imports for pyjanitor's dataframe accessors."""
-from .data_description import DataDescription
+"""Miscellaneous mathematical operators.
+
+Lazy loading used here to speed up imports.
+"""
+
+import warnings
+from typing import Tuple
+
+
+import lazy_loader as lazy
+
+scipy_special = lazy.load("scipy.special")
+ss = lazy.load("scipy.stats")
+pf = lazy.load("pandas_flavor")
+pd = lazy.load("pandas")
+np = lazy.load("numpy")
+pdtypes = lazy.load("pandas.api.types")

janitor/accessors/data_description.py

@@ -1,6 +1,8 @@
+import lazy_loader as lazy
 from typing import Dict, List, Union
-import pandas_flavor as pf
-import pandas as pd
+
+pf = lazy.load("pandas_flavor")
+pd = lazy.load("pandas")
 
 
 @pf.register_dataframe_accessor("data_description")

janitor/functions/impute.py

@@ -1,13 +1,15 @@
 """Implementation of `impute` function"""
 from typing import Any, Hashable, Optional
 
+import lazy_loader as lazy
 import numpy as np
 import pandas_flavor as pf
 import pandas as pd
-from scipy.stats import mode
 
 from janitor.utils import deprecated_alias
 
+ss = lazy.load("scipy.stats")
+
 
 @pf.register_dataframe_method
 @deprecated_alias(column="column_name")
@@ -100,7 +102,7 @@ def impute(
         "mean": np.mean,
         "average": np.mean,  # aliased
         "median": np.median,
-        "mode": mode,
+        "mode": ss.mode,
         "minimum": np.min,
         "min": np.min,  # aliased
         "maximum": np.max,

janitor/math.py

@@ -1,16 +1,20 @@
-""" Miscellaneous mathematical operators. """
+"""Miscellaneous mathematical operators.
+
+Lazy loading used here to speed up imports.
+"""
 
 import warnings
 from typing import Tuple
 
-import numpy as np
-import pandas as pd
-import pandas_flavor as pf
-from pandas.api.types import is_numeric_dtype
-from scipy.special import expit
-from scipy.special import logit as scipy_logit
-from scipy.special import softmax as scipy_softmax
-from scipy.stats import norm
+
+import lazy_loader as lazy
+
+scipy_special = lazy.load("scipy.special")
+ss = lazy.load("scipy.stats")
+pf = lazy.load("pandas_flavor")
+pd = lazy.load("pandas")
+np = lazy.load("numpy")
+pdtypes = lazy.load("pandas.api.types")
 
 
 @pf.register_series_method
@@ -94,7 +98,7 @@ def sigmoid(s: pd.Series) -> pd.Series:
     :param s: Input Series.
     :return: Transformed Series.
     """
-    return expit(s)
+    return scipy_special.expit(s)
 
 
 @pf.register_series_method
@@ -125,7 +129,7 @@ def softmax(s: pd.Series) -> pd.Series:
     :return: Transformed Series.
     """
 
-    return pd.Series(scipy_softmax(s), index=s.index, name=s.name)
+    return pd.Series(scipy_special.softmax(s), index=s.index, name=s.name)
 
 
 @pf.register_series_method
@@ -168,7 +172,7 @@ def logit(s: pd.Series, error: str = "warn") -> pd.Series:
         else:
             pass
     s[outside_support] = np.nan
-    return scipy_logit(s)
+    return scipy_special.logit(s)
 
 
 @pf.register_series_method
@@ -188,7 +192,7 @@ def normal_cdf(s: pd.Series) -> pd.Series:
     :param s: Input Series.
     :return: Transformed Series.
     """
-    return pd.Series(norm.cdf(s), index=s.index)
+    return pd.Series(ss.norm.cdf(s), index=s.index)
 
 
 @pf.register_series_method
@@ -229,7 +233,7 @@ def probit(s: pd.Series, error: str = "warn") -> pd.Series:
             pass
     s[outside_support] = np.nan
     with np.errstate(all="ignore"):
-        out = pd.Series(norm.ppf(s), index=s.index)
+        out = pd.Series(ss.norm.ppf(s), index=s.index)
     return out
 
 
@@ -313,7 +317,7 @@ def ecdf(s: pd.Series) -> Tuple[np.ndarray, np.ndarray]:
     :raises TypeError: if series is not numeric.
     :raises ValueError: if series contains nulls.
     """
-    if not is_numeric_dtype(s):
+    if not pdtypes.is_numeric_dtype(s):
         raise TypeError(f"series {s.name} must be numeric!")
     if not s.isna().sum() == 0:
         raise ValueError(f"series {s.name} contains nulls. Please drop them.")

janitor/utils.py

@@ -1,14 +1,23 @@
-"""Miscellaneous internal PyJanitor helper functions."""
+"""Miscellaneous mathematical operators.
+
+Lazy loading used here to speed up imports.
+"""
 
 import os
-import socket
 import sys
-from warnings import warn
-from functools import singledispatch, wraps
+import socket
 from typing import Callable, Dict, Iterable, Union
+from functools import singledispatch, wraps
+from warnings import warn
+
+import lazy_loader as lazy
 
-import numpy as np
-import pandas as pd
+scipy_special = lazy.load("scipy.special")
+ss = lazy.load("scipy.stats")
+pf = lazy.load("pandas_flavor")
+pd = lazy.load("pandas")
+np = lazy.load("numpy")
+pdtypes = lazy.load("pandas.api.types")
 
 
 def check(varname: str, value, expected_types: list):