minimal check for import V3

[guihao-liang]

Branched from #3152.

As it is discussed with @hoytak, @TobyRoseman, and @nickjong in #3156, I provide a simpler version for minimal package lazy import in order to lower the maintenance cost.

---

update 07/05/2020

This PR aims to provide some instructions to users who install the minimal package but somehow wind up using some ML functionalities.

For example, minimal pkg doesn't have TF installed as its dependency. When a user uses ML toolkit that relies on TF, he/she should see the instruction to install the full package's dependency.

Traceback (most recent call last):
...
ModuleNotFoundError: No module named 'tensorflow'. This is a minimal package for SFrame only, without tensorflow pinned as a dependency. You can try to install all required packages by installing the full package by:
**pip install turicreate==6.2**

Instead of letting them install each package individually, let them install turicreate==VERSION to rely on the full package's requirement list to install all required dependencies.

[hoytak]

Nice! Let's move this function into _tf_utils as it's copy-pasted elsewhere, then there's only one of it.

[guihao-liang]

there must be a reason for it to exist everywhere before. If we need to clean it, why now?

[hoytak]

Didn't you add _lazy_import_tensorflow? It was added in this PR.

[guihao-liang]

Yeah, it's a function name to wrap the existing code.

[guihao-liang]

for 3 files (drawing, activity, and style), at top-level, I wrap

    import tensorflow.compat.v1 as _tf
    # This toolkit is compatible with TensorFlow V2 behavior.
    # However, until all toolkits are compatible, we must call `disable_v2_behavior()`.
    _tf.disable_v2_behavior()

into,

def _lazy_import_tensorflow():
    _tf = minimal_package_import_check("tensorflow.compat.v1")
    # This toolkit is compatible with TensorFlow V2 behavior.
    # However, until all toolkits are compatible, we must call `disable_v2_behavior()`.
    _tf.disable_v2_behavior()
    return _tf

and only for the sound classifier, I wrap

    _utils.suppress_tensorflow_warnings()
    import tensorflow.compat.v1 as _tf
    # This toolkit is compatible with TensorFlow V2 behavior.
    # However, until all toolkits are compatible, we must call `disable_v2_behavior()`.
    _tf.disable_v2_behavior()

into

def _lazy_import_tensorflow():
    # Suppresses verbosity to only errors
    _utils.suppress_tensorflow_warnings()

    _tf = minimal_package_import_check("tensorflow.compat.v1")
    # This toolkit is compatible with TensorFlow V2 behavior.
    # However, until all toolkits are compatible, we must call `disable_v2_behavior()`.
    _tf.disable_v2_behavior()

    return _tf

These should not be considered as the duplicate of code because each tool may have its own settings on tf. Leaving it at the beginning of the file will allow people to know what's the tf configuration more clearly. I only add 2 more lines around. I don't believe that these 2 more lines are considered as the liability I shouldn't put here since people may not be able to maintain it because these 2 lines increase the mental burden on them so that they can't understand it.

Second, if this is considered as the duplicate of code, we wrap all common imports into on function to remove duplicate of import tensorflow orimportosorimport sys`.

@hoytak how do you think?

[guihao-liang]

passed internal tests (112066).

[guihao-liang]

from #3180:

core implementation for lazy import (including comments): 392 lines added
client adoption code: 48 lines added

from this PR:
core implementation for lazy import (including comments): 103 lines added
client adoption code: 165 lines added

The team favors later one is reasonable. Even more labor work and more client code affected, this PR comes with lower risk and liability.

[hoytak]

Didn't you add _lazy_import_tensorflow? It was added in this PR.

[guihao-liang]

@hoytak @TobyRoseman updated in ed670ff, the version of turicreate is also modified to resolve the complaint that,

ERROR: <package> has requirement turicreate==6.1.1+minimal, but you'll have turicreate 6.1.1 which is incompatible.

[guihao-liang]

this 6d80701 will trim the "+minimal" from local version (the downstream version) to prompt the upstream version to user if they use any ML toolkits from the minimal version.

[hoytak]

Good work! Ready to go.