Skip to content

Latest commit

 

History

History
111 lines (86 loc) · 2.81 KB

google-colab.md

File metadata and controls

111 lines (86 loc) · 2.81 KB
Raw

Running on Google Colab

Google Colab is free hosted Jupyter environment.

Here are some tips for running data prep kit application code on Google Colab.

How to Determine if We are Running on Colab?

Use one of these code snippets

import os

if os.getenv("COLAB_RELEASE_TAG"):
   print("Running in Colab")
   RUNNING_IN_COLAB = True
else:
   print("NOT in Colab")
   RUNNING_IN_COLAB = False

Installing Dependencies

We need to install data prep kit and libraries in Colab

if RUNNING_IN_COLAB:
    ! pip install  --default-timeout=100  data-prep-toolkit-transforms-ray==0.2.1.dev3

Downloading Data Files on Colab

if RUNNING_IN_COLAB:
    !mkdir -p 'input'
    !wget -O 'input/1.pdf'  'remote_file_url'

Ray Runtime Settings

These are some recommended settings for running RAY based notebooks. You can use these as starting points and tweak for your application

if RUNNING_IN_COLAB:
  RAY_RUNTIME_WORKERS = 2
  RAY_NUM_CPUS =  0.3
  RAY_MEMORY_GB = 2  # GB

It is recommended to set cpu per worker (RAY_NUM_CPUS) to a low number. Otherwise Ray jobs seem to hang and will not complete.

Fuzzy Dedupe Settings

Start with the following settings before launching fuzzy dedupe job.

Here is the infrastructure section for fuzzy dedupe. Again we recommend to keep CPU share low.

    # infrastructure
    "fdedup_bucket_cpu": 0.3,
    "fdedup_doc_cpu": 0.3,
    "fdedup_mhash_cpu": 0.3,
    "fdedup_num_doc_actors": 1,
    "fdedup_num_bucket_actors": 1,
    "fdedup_num_minhash_actors": 1,
    "fdedup_num_preprocessors": 1,

Here is full code for completeness

local_conf = {
    "input_folder": input_folder,
    "output_folder": output_folder,
}
worker_options = {"num_cpus" : RAY_NUM_CPUS}

params = {
    # where to run
    "run_locally": True,
    # Data access. Only required parameters are specified
    "data_local_config": ParamsUtils.convert_to_ast(local_conf),
    # Orchestration parameters
    "runtime_worker_options": ParamsUtils.convert_to_ast(worker_options),
    "runtime_num_workers": RAY_RUNTIME_WORKERS,
    # columns used
    "fdedup_doc_column": "contents",
    "fdedup_id_column": "int_id_column",
    "fdedup_cluster_column": "hash_column",
    # infrastructure
    "fdedup_bucket_cpu": 0.3,
    "fdedup_doc_cpu": 0.3,
    "fdedup_mhash_cpu": 0.3,
    "fdedup_num_doc_actors": 1,
    "fdedup_num_bucket_actors": 1,
    "fdedup_num_minhash_actors": 1,
    "fdedup_num_preprocessors": 1,
    # fuzzy parameters
    "fdedup_num_permutations": 64,
    "fdedup_threshold": 0.7,
    "fdedup_shingles_size": 5,
    "fdedup_delimiters": " "
}

# Pass commandline params
sys.argv = ParamsUtils.dict_to_req(d=params)

launcher = RayTransformLauncher(FdedupRayTransformConfiguration())
return_code = launcher.launch()