from lib.data import mnist
from lib.fnn.shallowNet import ShallowNet
from lib.fnn.performance import trainTestNetwork
import lib.nonlinFunctions as nonlinFunctions
# Load data
dataMNIST = mnist.load("/home/aleksejs/Downloads/mnist_data/")
# Set parameters
param = {
# Network parameters
'netClass': ShallowNet,
'ny': 10, # Number of possible outputs (for binarization)
'nHid': [], # no hidden layers
'nonlinFunc': nonlinFunctions.func_relu,
'nonlinPrimFunc': nonlinFunctions.fprim_relu,
'bSTD': 10**-4, # variance of initial thresholds
'wSTD': 10**-4, # variance of initial weights
# Testing parameters
'etaPref': 10**-8, # Prefactor for learning rate
'nEpoch': 100, # Number of times to sweep the entire data
'nMini': 32, # Number of datapoints per minibatch
}
trainTestNetwork(*dataMNIST.values(), param)
import numpy as np
import matplotlib.pyplot as plt
from lib.data import mnist
from lib.autoenc.rbdeep import RaoBallardDeep
import lib.nonlinFunctions as nonlinFunctions
# Load data
dataMNIST = mnist.load("/home/alyosha/Downloads/mnist_data/")
nData = 5
dataIdxs = np.arange(nData)
xarr = dataMNIST['train_images'][dataIdxs].astype(float)
nPix = xarr.shape[1] * xarr.shape[2]
for i in range(nData):
xarr[i] /= np.linalg.norm(xarr[i])
sig = nonlinFunctions.func_id
sigp = nonlinFunctions.fprim_id
# Specify parameters
param = {
'nNode': [nPix, 5],
'dt': 0.1, # ms, timestep
'tauX': 1.0, # ms, neuronal timescale
'tauU': 100.0, # ms, plasticity timescale
'inputNoise': 0.0,
'uNorm': 1.0,
'nonlinFunc': [sig, sig],
'nonlinPrimFunc': [sigp, sigp]
}
# Run
from lib.data import mnist
mnistData = mnist.load("/home/aleksejs/Downloads/mnist_data/")
select_images = [0, 1, 15, 138, 2000]
mnist.plot(mnistData["train_images"], mnistData["train_labels"], select_images)