Build a Softmax classifier for fashion MNIST dataset
- Python3
- h5py==2.8.0
- matplotlib==2.1.1
- numpy==1.15.1
You can use pip install to install the packages you need.
For more detailed description, see my blog
To run this program
python train.py
It will print the training loss every 50 epochs. After the program finishes, a train loss curve and an accuracy curve will be plot automatically. Check output folder, the predicted_labels.h5 containing the labels of last 3000 test samples will be created after the program is finished.
To run this model with different parameters to repeat experiment results above, change the corresponding parameters in train.py from following functions
model = LogisticRegression(num_iters = 300,
alpha=0.8,
with_reg = True,
with_monentum = True)
results = model.train(train_data.T, train_label.T, Y, test_data_2000.T,test_label.T[0], print_loss=True)
To print out confusion matrix, uncomment these two lines in train.py
# cfn = cnf_matrix(label_prediction_test,test_label)
# print(cfn)