TF Train MNIST Data
The TF Train MNIST Data operator is an extension of the python27 operator. It runs tensorflow code that trains a simple neuronal network model to recognize numbers from a handwriting image.
The code is based on the tensorflow MNIST example https://www.tensorflow.org/tutorials/layers
Configuration Parameters
|
Parameter |
Type |
Description |
|---|---|---|
| modeldir | string | The command-line parameter that specifies where the model is
stored in the repository.
Default: "/modeldir" |
| target_file | string | Contains the filename of the tar-artifact. |
| train_epochs | int | Split training steps into batches.
Default: 100000 |
| batch_size | int | Size of these batches.
Default: 100 |
| demo | bool | Introduces additional time between each training.
Default: true |
| stochastic |
bool |
Using small batches of random data is called stochastic
training.
Default: true |
Input
|
Input |
Type |
Description |
|---|---|---|
|
inSampleDataDir |
string |
The file path to the training data. |
Output
|
Output |
Type |
Description |
|---|---|---|
|
outBlobPath |
string |
The file path to the produced model tar-artifact. |