Add image classification notebook #1
Conversation
| demo = True | ||
|
|
||
| stop_criterion = { | ||
| 'time_limits': time_limits, |
There was a problem hiding this comment.
Let's not put the time_limits here since we don't yet support it in the backend.
Same goes for max_metric.
There was a problem hiding this comment.
Sure, WIP we would have the implementation soon.
There was a problem hiding this comment.
Yes. But until we have it, can we remove this ?
|
|
||
| # type of net_list is ag.space.List | ||
|
|
||
| # method 1 (complex but flexiable): specify the net_list using get_model |
There was a problem hiding this comment.
Can you remove the commented out part for method 1 ?
|
|
||
| # method 1 (complex but flexiable): specify the optim_list using get_optim | ||
| # optimizers = ag.Optimizers([ag.optim.get_optim('sgd'), | ||
| # ag.optim.get_optim('adam')]) |
| Example 2: Amazon Internal Image | ||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
|
||
| We can also access `Amazon Image assist <https://image-assist.amazon.com>`__ |
There was a problem hiding this comment.
This can easily slip the cracks and become public documentation once we open source this. I'd recommend not to put an internal Amazon link in here.
| get_model('cifar_resnet20_v1'), | ||
| get_model('cifar_resnet56_v1'), | ||
| get_model('cifar_resnet110_v1')]), | ||
| get_model('resnet18_v1'), |
There was a problem hiding this comment.
So for now are we going with just 2 default models ?
There was a problem hiding this comment.
We can support any models. The point is to now using cifar pre-trained models for cifar dataset.
There was a problem hiding this comment.
Will the end user dataset be CIFAR only ? Changing defaults affects the end user use case as well.
docs/source/install.rst
Outdated
|
|
||
| .. note:: | ||
|
|
||
| The whl would be updated soon. |
There was a problem hiding this comment.
Minor nit : wheel instead of whl
| max_num_gpus = 1 | ||
| max_num_cpus = 4 | ||
| max_training_epochs = 2 | ||
| demo = True |
There was a problem hiding this comment.
What is this argument doing?
There was a problem hiding this comment.
They are used in stop_criterion and resources_per_trial.
There was a problem hiding this comment.
I meant the demo argument, how is that used in stop_criterion?
| max_metric = 0.80 | ||
|
|
||
| stop_criterion = { | ||
| 'time_limits': time_limits, |
There was a problem hiding this comment.
Can we remove time_limits and max_metric since they are not implemented?
docs/source/install.rst
Outdated
| @@ -8,6 +8,17 @@ Installation | |||
| git clone https://github.com/awslabs/auto-ml-with-gluon.git && cd auto-ml-with-gluon | |||
| python setup.py install | |||
|
|
|||
| .. note:: | |||
|
|
|||
| Before installing from source, please have the MXNet and python installed. | |||
There was a problem hiding this comment.
Is there a particular minimum MXNet version they need to install ?
If yes, let's list it out here.
| @@ -0,0 +1,275 @@ | |||
| """1. Create Searcher and Scheduler | |||
There was a problem hiding this comment.
just a minor comment. You may use git mv src target to move the file and keep the git history.
There was a problem hiding this comment.
Sure, feel free to delete this one.
|
close inactive PR |
* introduce compile_model to tabular predictor (cherry picked from commit eeb875b) * separate compile method from fit and save, and addressed some review comments * fix compile function * build onnx model without zipmap * improve robustness * address review comments and put compile_time into part of model_info * replace _input_types_post_process with _features * fix compile_time in leaderboard * bug fix in model_info * add comments * compile base model via getting its ancestors from compile_models interface * move model-specific logic outside of abstract_model * move model-specific logic outside of abstract_model * Refactored compiler API and logic (#1) Co-authored-by: Nick Erickson <neerick@amazon.com>
Description of changes: