4. Parallel computing
library that scales
the existing Python
ecosystem : NumPy,
Pandas , Scikit
PARALLELIZATION FLEXIBLE RESPONSIVE
Task scheduling
interface for more
custom workloads
and integration.
Single machine or
distributed
workloads across
GCP clusters
Runs resiliently up
to 1000s cores.
Horizontal and
vertical scaling.
6. 6
Example:
1. Dask in GKE cluster
2. Models deployment (Helm)
3. Faster training and
workload parallelization
4. Scale via Kubernetes or
Dask (threads/workers)
Demo: simple tree
summation computation
with Dask installed in our
cluster.
9. WHY DASK IS
GOOD FOR US
OPEN SOURCE
Our setup is on GKE
with a single Dask
cluster deployed
with Helm. Open
Source solution
configurable for all
the ML projects.
SCALING
We scale the number
of workers (threads)
dividing the
workload in multiple
units. Horizontal or
vertical scaling via
dashboard.
SPEED
We speed up the
computation
achieving faster
times in training
models
10. Next?
Creation of Dask clusters
by computation context,
saving costs .
Storage and dataframes
parallelization refactoring
the model
implementation.
12. CREDITS
This is where you give credit to the ones who are part of this
project.
Did you like the resources on this template? Get them for
free at our other websites.
◂ Presentation template by Slidesgo
◂ Icons by Flaticon
◂ Infographics by Freepik
◂ Images created by Freepik and rawpixel -Freepik
◂ Author introduction slide photo created by Freepik
◂ Text & Image slide photo created by Freepik.com