Feat/Add Multi Layer Perceptron Demo code #18
Conversation
jcabrero
force-pushed
the
feat/add_mlp_demo
branch
from
1ffffeb to
bda7c50
Compare
| # Multi-Layer Perceptron Demo | ||
| **This folder was generated using `nada init`** | ||
|
|
||
| To execute this tutorial, you may potentially need to install the `requirements.txt` appart from nada-ai: |
There was a problem hiding this comment.
nit: typo appart -> apart
| @@ -0,0 +1,7 @@ | |||
| name = "text_classification" | |||
There was a problem hiding this comment.
nit: name -> multi_layer_perceptron
| @@ -0,0 +1,7 @@ | |||
| scikit-learn~=1.4.2 | |||
There was a problem hiding this comment.
nit: do we need sklearn, pandas, requests and matplotlib here?
There was a problem hiding this comment.
matplotlib is required for graphs that are plot on the demo
| @@ -116,7 +116,7 @@ def __init__(self) -> None: | |||
| self.linear_1 = torch.nn.Linear(4, 2) | |||
| self.relu = torch.nn.ReLU() | |||
|
|
|||
| def forward(self, x: na.NadaArray) -> na.NadaArray: | |||
| def forward(self, x: np.ndarray) -> np.ndarray: | |||
There was a problem hiding this comment.
nit: these arrays are torch.tensor instead of np.ndarray, no?
| requests~=2.31.0 | ||
| torchvision~=0.18.1 |
There was a problem hiding this comment.
nit: do we need torchvision and pillow here?
| @@ -0,0 +1,497 @@ | |||
| { | |||
There was a problem hiding this comment.
(just gonna comment on first line b/c notebooks x github is annoying)
we're doing inference_result[...] / (2 ** na.get_log_scale()) --> could do rational_to_float instead?
There was a problem hiding this comment.
Also: that first cell is insane. Can only imagine the pain you had to go through to come up with that fix - my condolences
| @@ -0,0 +1,799 @@ | |||
| { | |||
There was a problem hiding this comment.
btw just saw the confusion matrix --> our trained model is basically return 1
we might wanna look into addressing that class imbalance
There was a problem hiding this comment.
Agreed. Most medical examples are quite imbalanced. The original AI demo was balanced though. Maybe we could take it from there.
|
@jcabrero general comment: I think (correct me if i'm wrong?) this is our most demanding example on the network. |
Probably. However, in Nillion-Devnet it executes on a similar efficiency than other examples (same execution time). |
This adds the multi layer perceptron code.
EXTRA: Linting and formating code.