The goal is to write a Framework from scratch using only basic python tools and libraries
I unapologetically reinvent the best parts from the best sources
from raml.models import Sequential
from raml.layers import Dense
from raml.activations import LeakyRelu
from raml.costs import MSE
from raml.metrics import RMSE
from raml.utils import format_data, plot_history
from raml.preprocessing import Normalizer
from raml.datasets.load import Boston_House_Price
X, Y = Boston_House_Price()
(x_train, x_val), (y_train, y_val) = train_test_split(X, Y=Y, ratio=[0.7, 0.3], shuffle=True)
normalizer = Normalizer()
x_train = normalizer.fit(x_train)
x_val = normalizer.apply(x_val)
def train_model():
ITERATIONS = 1000
model = Sequential([
Dense(size=100, input_shape=X.shape, activation=LeakyRelu),
Dense(size=20, activation=LeakyRelu),
Dense(size=20, activation=LeakyRelu),
Dense(size=1, activation=Identity),
])
model.compile( cost = MSE(), metrics = [RMSE()] )
history = model.fit(x_train, y_train, epochs=ITERATIONS, x_val=x_val, y_val=y_val)
plot_history(history)
train_model()
Works beautifuly!
But wait, there is more! Checkout main.ipynb
for the latest example of tackling the MNIST dataset with 80% accuracy!!
It will only get better!