ModuleNotFoundError: No module named 'tfx.utils.dsl_utils'

Question

I did install tfx version 1.2.1 Python: 3.8.2 Tensorflow: 2.5.2 pip: 21.3.1 I use window and installed the package through pip.

The error occurred when I did:

import os
from tfx.components import CsvExampleGen
from tfx.utils.dsl_utils import external_input
base_dir = os.getcwd()
data_dir = os.path.join(os.pardir, "data")
examples = external_input(os.path.join(base_dir, data_dir))
example_gen = CsvExampleGen(input=examples)
context.run(example_gen)

The error:

ModuleNotFoundError                       Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_14604/719215164.py in <module>
      1 import os
      2 from tfx.components import CsvExampleGen
----> 3 from tfx.utils.dsl_utils import external_input
      4 base_dir = os.getcwd()
      5 data_dir = os.path.join(os.pardir, "data")

ModuleNotFoundError: No module named 'tfx.utils.dsl_utils'

I did full installation of tfx in which all packages are compatible. Any help is appreciated.

Answer 1

Instead of using 'external_input', send data_dir directly into the CsvExampleGen.

import os
from tfx.components import CsvExampleGen
base_dir = os.getcwd()
data_dir = os.path.join(os.pardir, "data")
example_gen = CsvExampleGen(input_base='data_dir')

This has worked for me. Looks like in tfx version 1._ , the module tfx.utils.dsl_utils doesn't exist.

Answer 2

Just do the below:

# Example 1
context= InteractiveContext()
from tfx.components import CsvExampleGen
example_gen = CsvExampleGen(input_base='data')
context.run(example_gen)

data is the folder that you would have downloaded from the author's github. Just specify the folder name directly (or the path to the data folder).

Answer 3

Try this instead

example_gen = CsvExampleGen(input_base=_data_root)
context.run(example_gen)

and change the Train to Split-train