Profile Earth Engine computation

The Earth Engine API provides tools for profiling the performance of your computations but they are not always the easiest to use to get the number you are looking for. The geetools library supercharge the original profiler to make any computation evaluation the easiest possible.

Set up environment

Install all the requireed libs if necessary. and perform the import satements upstream.

# uncomment if installation of libs is necessary
# !pip install earthengine-api geetools

import ee
import geetools
import pandas as pd

# uncomment if authetication to GEE is needed
# ee.Authenticate()
# ee.Intialize(project="<your_project>")

Example data

The following examples rely on a ee.FeatureCollection composed of three ecoregion features that define regions by which to reduce image data. The Image data are PRISM climate normals, where bands describe climate variables per month; e.g., July precipitation or January mean temperature.

ecoregions = (
    ee.FeatureCollection("projects/google/charts_feature_example")
    .select(["label", "value","warm"])
)

normClim = ee.ImageCollection('OREGONSTATE/PRISM/Norm91m').toBands()

default profiler

The default profiler from Earth Engine can be called as a context manager, it will print at the end of the cell the extensive description of your computation.

with ee.profilePrinting():
    normClim.geetools.byBands(
        regions = ecoregions,
        reducer = "mean",
        scale = 500,
        regionId = "label",
        bands = [f"{i:02d}_tmean" for i in range(1,13)],
    ).getInfo()

 EECU·s PeakMem Count  Description
  0.422     59k     6  Algorithm Image.reduceRegions
  0.236    568k    86  Loading assets: projects/google/charts_feature_example
  0.195    382k   951  (plumbing)
  0.136    121k   252  Loading assets: OREGONSTATE/PRISM/Norm91m/(...)
  0.042    1.1M    86  no description available
  0.006     200    72  Reprojecting pixels from GEOGCS["GCS_North_American_1983",DATUM["North_American_Datum_1983",SPHEROID[...] to GEOGCS["GCS_North_American_1983",DATUM["North_American_Datum_1983",SPHEROID[...]
  0.002    9.7k    15  Algorithm ImageCollection.toBands
  0.001    5.6k    15  Algorithm Image.rename
  0.001    3.8k    15  Algorithm Image.select
  0.001    3.2k    13  Algorithm Collection.reduceColumns with reducer Reducer.toList
  0.001    3.0k    14  Algorithm ReduceRegions.AggregationContainer
  0.000     384     3  Listing collection
  0.000    113k     3  Computing image mask from geometry
   -       121k    50  Loading assets: OREGONSTATE/PRISM/Norm91m
   -        90k    26  Algorithm Collection.reduceColumns
   -        44k    19  Algorithm List.map
   -        27k    14  Algorithm Dictionary.fromLists
   -        23k    40  Algorithm AggregateFeatureCollection.array
   -        11k    15  Algorithm Collection.loadTable
   -       9.5k    15  Algorithm ImageCollection.load
   -       8.6k     4  Algorithm Projection
   -       8.3k     1  Algorithm (user-defined function)
   -       6.9k     4  Algorithm ReduceRegions.ReduceRegionsEnumerator
   -       5.1k    15  Algorithm Collection.map
   -       3.1k    14  Algorithm Feature.select
   -       3.1k    10  Algorithm If
   -       3.1k     4  Algorithm Reducer.forEach
   -       2.9k    10  Algorithm Number.eq
   -       2.9k     4  Algorithm String.compareTo
   -       2.8k    37  Algorithm String
   -       2.8k    10  Algorithm ObjectType
   -       2.6k    20  Loading assets: OREGONSTATE/PRISM
   -       2.6k    20  Loading assets: projects/google
   -       1.7k     1  Algorithm Number.format
   -        440     5  Algorithm Reducer.mean
   -        240     7  Expression evaluation
   -         64    72  Algorithm Image.load computing pixels

This result is extremely useful but cannot be further explored in the notebook.

geetools profiler

The geetools profiler is a context manager object that fill a dictionary member (profile) with the content of the string profile. This dictionary can be transformed into a table easily.

# example with a simple function
with ee.geetools.Profiler() as p:
    ee.Number(3.14).add(0.00159).getInfo()
p.profile

{'EECU-s': [0.001, None],
 'PeakMem': [3120, 2900],
 'Count': [3, 3],
 'Description': ['(plumbing)', 'Algorithm']}

With a bigger method we can valorized the results as a pandas dataframe and extract key informations.

with ee.geetools.Profiler() as p:
    normClim.geetools.byBands(
        regions = ecoregions,
        reducer = "mean",
        scale = 500,
        regionId = "label",
        bands = [f"{i:02d}_tmean" for i in range(1,13)],
    ).getInfo()
df = pd.DataFrame(p.profile)
df.head()

	EECU-s	PeakMem	Count	Description
0	0.280	59000	6	Algorithm
1	0.246	567000	84	Loading
2	0.176	355000	943	(plumbing)
3	0.117	212000	252	Loading
4	0.041	1000000	86	no

# total EECU cost of the computation
float(df["EECU-s"].sum())

0.87