`import ecoscope` takes too long

[cisaacstern]

Currently importing ecoscope takes ~6 seconds.

$ time python3 -c "import ecoscope"
python3 -c "import ecoscope"  6.07s user 0.94s system 80% cpu 8.673 total

If I comment out this line

https://github.com/wildlife-dynamics/ecoscope/blob/b3b9d1543eee5357ef9b2df8a4135d47d40f39d4/ecoscope/__init__.py#L1

that is reduced to 0.3 seconds:

$ time python3 -c "import ecoscope"
python3 -c "import ecoscope"  0.03s user 0.01s system 73% cpu 0.060 total

IMHO 6 seconds is too long to wait for ecoscope to import. An example of why this matters is in test development: if test session takes a minimum of 6 seconds to initialize, that adds up to a lot of developer time over the course of hundreds or thousands of test invocations.

I am going to see if I can reduce import time and open a PR for review 😃

[cisaacstern]

A little more detail on import time of the various components:

• from ecoscope import analyis ➡️ 4s
• from ecoscope import base ➡️ circular import w analysis, can't be imported on it's own (side note: this seems like its own issue, i would think things named base don't have other dependencies within the package)
• from ecoscope import contrib ➡️ 0.2s
• from ecoscope import io ➡️ 0.98s
• from ecoscope import mapping ➡️ 0.58s
• from ecoscope import plotting ➡️ 1.92s

4 + 0.2 + 0.98 + 0.58 + 1.92 = 7.68 seconds (so these numbers check out)

[cisaacstern]

I am going to see if I can reduce import time and open a PR for review 😃

Note, for now to unblock development of #141, I've just commented out this line:

https://github.com/wildlife-dynamics/ecoscope/blob/818696e5ad69e216c902ac73e180bcd0753588b2/ecoscope/__init__.py#L1-L2

I can revisit a more general fix soon!

[cisaacstern]

Deferring import of mapclassify saves about 1.5s

diff --git a/ecoscope/analysis/speed.py b/ecoscope/analysis/speed.py
index 5873b72..825da98 100644
--- a/ecoscope/analysis/speed.py
+++ b/ecoscope/analysis/speed.py
@@ -1,7 +1,6 @@
 import typing

 import geopandas as gpd
-import mapclassify
 import pandas as pd
 import shapely

@@ -64,7 +63,16 @@ def apply_classification(x, k, cls_method="natural_breaks", multiples=[-2, -1, 1
     multiples : Listlike
         The multiples of the standard deviation to add/subtract from the sample mean to define the bins. defaults=
     """
-
+    import mapclassify
+
+    classification_methods = {
+        "equal_interval": mapclassify.EqualInterval,
+        "natural_breaks": mapclassify.NaturalBreaks,
+        "quantile": mapclassify.Quantiles,
+        "std_mean": mapclassify.StdMean,
+        "max_breaks": mapclassify.MaximumBreaks,
+        "fisher_jenks": mapclassify.FisherJenks,
+    }
     classifier = classification_methods.get(cls_method)
     if not classifier:
         return
@@ -82,12 +90,3 @@ default_speed_colors = [
     "#fc8d59",
     "#d73027",
 ]
-
-classification_methods = {
-    "equal_interval": mapclassify.EqualInterval,
-    "natural_breaks": mapclassify.NaturalBreaks,
-    "quantile": mapclassify.Quantiles,
-    "std_mean": mapclassify.StdMean,
-    "max_breaks": mapclassify.MaximumBreaks,
-    "fisher_jenks": mapclassify.FisherJenks,
-}
~
~
~

[atmorling]

I've tested the following:

• Moving the functions analysis/astronomy.py into base/utils.py to solve the circular import issue
• Deferring expensive imports like mapclassify and astroplan
• Rearranging the imports in ecoscope/__init__.py so that analysis is loaded last

This reduces the import time to around ~2.7 seconds.

Deferring the load of ecoscope.analysis.UD and ecoscope.mapping (removing their imports from the relevant __init__.pys) gets this down to ~1.3 seconds. This means one would have to explicitly import these modules in addition to import ecoscope .

Deferring the expensive imports isn't exactly PEP8 Approved^TM, so I think playing around with lazy imports is worthwhile.

Also learned that tuna for visualizing import profiler outputs. (See the pretty screenshot)

[cisaacstern]

amazing progress!