udacity / ml_sagemaker_studies Goto Github PK

In Moon_Classification_Solution.ipynb, the original code below would cause an error ValueError: framework_version or py_version was None, yet image_uri was also None. Either specify both framework_version and py_version, or specify image_uri. So I specified py_version='py3', cause the framework version only supports py2 and py3, which fixed the problem. Or I guess just add !pip install sagemaker==1.72.0 like notebooks in another repo would also solve the issue.

# import a PyTorch wrapper
from sagemaker.pytorch import PyTorch

# specify an output path
# prefix is specified above
output_path = 's3://{}/{}'.format(bucket, prefix)

# instantiate a pytorch estimator
estimator = PyTorch(entry_point='train.py',
                    source_dir='source_solution', # this should be just "source" for your code
                    role=role,
                    framework_version='1.0',
                    py_version='py3', ### <------------------------ added a line here
                    train_instance_count=1,
                    train_instance_type='ml.c4.xlarge',
                    output_path=output_path,
                    sagemaker_session=sagemaker_session,
                    hyperparameters={
                        'input_dim': 2,  # num of features
                        'hidden_dim': 20,
                        'output_dim': 1,
                        'epochs': 80 # could change to higher
                    })

Hi,

Sorry, but pandas.DataFrame.as_matrix() appears to be deprecated. Can this notebook please be updated as a result?

def train_test_split(transaction_df, train_frac= 0.7, seed=1):

...

    # convert the df into a matrix for ease of splitting
    df_matrix = transaction_df.as_matrix()
...

Thank you!

There is a small bug in the 1_Data_Exploration notebook. The line:
% matplotlib inline
should be:
%matplotlib inline
I open this Issue and not a PR because of git problems tracking ipynb files.

Hello! Has anyone had the same problem with content_type parameter from deploy method in the Time_Series_Forecasting project?

Hi, is the Ukrainian language supported ?
if not, is it possible train model for UA lang?

tnx

In project 2 Notebook3 has some redundant line

In the time series forecasting project
https://github.com/udacity/ML_SageMaker_Studies/blob/master/Time_Series_Forecasting/Energy_Consumption_Exercise.ipynb

we have a function make_time_series that fails here
index = pd.DatetimeIndex(start=t_start, end=t_end, freq=freq)

Instead this should be
index = pd.date_range(start=t_start, end = t_end, freq= freq)

I'm using pandas version 1.0.3

Thanks

Tony