Chris Pollett > Students >
Swathi
    ( Print View)

    [Bio]

    [Blog]

    [C297 Proposal]

    [D1:Dataset Overview - pdf(ipynb)]

    [Deliverable 1]

    [Deliverable 2]

    [Visualization Techniques - pdf(Slides)]

    [Deliverable 3]

    [Deliverable 4]

    [CS 297 Report PDF]

    [C298 Proposal]

Deliverable 4:Neural Network Classification


Description:

In the previous deliverables we saw Classification, Clustering methods as well as Binary Classification methods to understand and analyze the Heart data of Tabula Sapiens dataset. In this deliverable, I built a Neural Network Classifier to classify the different types of cells present in Heart data and verified how accurately it predicted the different cells.

Implementation Steps:

  1. Download and install Python, Anaconda and Jupyer Notebook.
  2. Download Tabula Sapiens - Heart Dataset from CZ Biohub website.
  3. Install anndata, scanpy, numpy, pandas, matplotlib, seaborn python packages.
  4. Execute the classification and clustering code snippets one after the other as mentioned in Juputer Notebook(ipynb) file.

Results:

  1. The Heart data i.e genes have been converted to floating point data and the categorical variable cell type has been label encoded to supply as inputs to the Neural Network model.

    2. Neural network Xtrain Neural network Ytrain
  2. The Neural Network Classifier contains 2 hidden layers with 128 neurons and 64 neurons, the activation function is ReLU. The input is the features of Heart data. The output layer contains 6 neurons to predict the cell type with softmax activation function.

    3. Neural network model
  3. The model was trained for 10 epochs. We got an impressive accuracy of 98.6%.

    4. Neural network accuracy