def visualize_neurons():
input_neurons = request.form['input_neurons']
hidden_neurons = request.form['hidden_neurons']
num_input = int(input_neurons)
num_hidden = int(hidden_neurons)
xs_ys, weights = plot_net(num_input, num_hidden, 480, 600)
db.create_key('data', xs_ys)
db.create_key('weights', weights)
db.delete_key('next_weight_layer')
return render_template('visualize_neurons.html',
data=db.read_key('data'),
weights=db.read_key('weights'))
def visualize_data():
# Get in number of input neurons
num_input = int(request.form['input_neurons'])
num_hidden = int(request.form['hidden_neurons'])
# Generate data - updates database with data.
df = generate_data(num_input)
# Inside plot net, get one row from the dataFrame
coords, weights = plot_net(num_input, num_hidden, 480, 600)
# Add something to the values so it doesn't plot zeros.
layer1 = Linear(n_in=num_input,
n_out=num_hidden,
activation_function=sigmoid)
layer2 = Linear(n_in=num_hidden, n_out=1, activation_function=sigmoid)
nn = TwoLayerNetwork(layers=[layer1, layer2], random_seed=823)
# Sample one row of x data
df_x = db.read_key('dataframe')[db.read_key('x_cols')]
samp = np.random.randint(0, db.read_key('num_obs'))
x_row = np.array(df_x.iloc[samp, :], ndmin=2)
# Sample one y observation
# import pdb; pdb.set_trace()
df_y = db.read_key('dataframe')["Y"]
y_row = np.array(df_y.iloc[samp], ndmin=2)
# Take this single observation and run it through the network to generate
# values for the layers
layer_values, prediction = nn.return_forwardpass_info(x_row)
print("layer_values", layer_values)
layers_clean = [clean_layer(layer) for layer in layer_values]
# Simply extract the weight values
weight_values = nn.return_weights()
weight_values = clean_weights(weight_values)
# In the coords and weights "objects", update the weights and coordinates
# values
coords = update_coordinate_values(layers_clean, coords)
weights = update_weight_values(weights, weight_values)
# Compute loss
loss = nn.loss(prediction, y_row)
db.create_key("coordinates", coords)
db.create_key("weights", weights)
db.create_key("x_data", x_row)
db.create_key("layer_values", layer_values)
db.create_key("prediction", prediction)
db.create_key("next_grad", loss)
db.create_key("neural_net", nn)
db.create_key("next_layer", len(nn.layers))
# Render the template!
return render_template('visualize.html',
data=db.read_key("coordinates"),
weights=db.read_key("weights"))
def update_next_weights():
print("Next layer:", db.read_key('next_layer'))
if db.read_key('next_layer') < 1:
weights = db.read_key('weights')
return jsonify(data=weights)
nn = db.read_key("neural_net")
next_grad = db.read_key("next_grad")
weights_before = db.read_key('weights')
weight_values_before_clean = nn.update_num_layer_weights(next_grad, num=1)
weight_values_after_clean = clean_weights(weight_values_before_clean)
weights = update_weight_values(db.read_key('weights'),
weight_values_after_clean)
db.update_key('weights', weights)
time.sleep(1)
print("Updating weights before layer", db.read_key('next_layer') + 1)
return jsonify(data=weights)
def update_next_neurons():
print("Next layer:", db.read_key('next_layer'))
if db.read_key('next_layer') > len(db.read_key('neural_net').layers):
coords = db.read_key('coordinates')
return jsonify(data=coords)
nn = db.read_key("neural_net")
if db.read_key('next_layer') == 0:
x_row = read_x_row()
db.update_key("x_data", x_row)
else:
x_row = db.read_key("x_data")
layers = nn.return_num_layers(x_row)
layers_clean = [clean_layer(layer) for layer in layers]
print("layers_clean", layers_clean)
coords = update_coordinate_values(layers_clean, db.read_key('coordinates'))
print("Coordinates:", coords)
db.update_key('coordinates', coords)
time.sleep(5)
print("Updated neurons in layer", db.read_key('next_layer') - 1)
return jsonify(data=coords)
def reset_data():
print("The reset data button was clicked")
db.delete_all_keys()
if not db.read_key("data"):
db.create_key("data", [])
if not db.read_key("weights"):
db.create_key("weights", [])
if not db.read_key("coordinates"):
db.create_key("coordinates", [])
if db.read_key("next_weight_layer"):
db.delete_key("next_weight_layer")
return render_template('index.html',
coordinates=db.read_key('coordinates'),
data=db.read_key('data'),
weights=db.read_key('weights'))