0

I'm doing some identification with self organizing map & python. I use a raster data and I convert it to a numpy array. I want to visualize it with matplotlib but it shows an error like this.

And here is the python code.

from __future__ import division
import gdal

import numpy as np
import matplotlib.pyplot as plt 
from matplotlib.patches import Rectangle


fn = "D:/kuliah/bismillahsatu/djengkol/ndwigabung.tif"
ds = gdal.Open(fn)

band1 = ds.GetRasterBand(1).ReadAsArray()

#print(band1.shape)
#print(band1)

network_dimensions = np.array([5,5])
n_iterations = 10000
init_learning_rate = 0.01

normalise_data = True

#if True, assume all data is on common scale
#if false, normalise to [0 1] range along each column
normalise_by_column = False 

#establish size variables based ondata
m = band1.shape[0]
n = band1.shape[1]

#initial neighborhood radius
init_radius = max(network_dimensions[0], network_dimensions[1]) / 2

#radius decay parameter
time_constant = n_iterations / np.log(init_radius)

#we want to keep a copy of the raw data for later
data = band1


#check if data needs to be normalised
if normalise_data:
    if normalise_by_column:
        #normalise along each column
        col_maxes = band1.max(axis=0)
        data = band1/col_maxes[np.newaxis, :]
    else:
        #normalise entire dataset
        data = band1 / data.max()

net = np.random.random((network_dimensions[0], network_dimensions[1],m))

def find_bmu (t,net,m):

    bmu_idx = np.array([0,0])
    min_dist = np.iinfo(np.int).max
    for x in range(net.shape[0]):
        for y in range(net.shape[1]):
            w = net[x, y, :].reshape(m,1)
            sq_dist = np.sum((w - t) ** 2)
            if sq_dist < min_dist:
                min_dist = sq_dist
                bmu_idx = np.array([x,y])
    bmu = net[bmu_idx[0], bmu_idx[1], :].reshape(m,1)
    return (bmu, bmu_idx)

def decay_radius(initial_radius, i, time_constant):
    return initial_radius * np.exp(-i / time_constant)

def decay_learning_rate(initial_learning_rate,i, n_iterations):
    return initial_learning_rate * np.exp(-i / n_iterations)

def calculate_influence(distance, radius):
    return np.exp(-distance /(2*(radius**2)))

for i in range (n_iterations):

    t = data[:, np.random.randint(0, n)].reshape(np.array([m,1]))
    bmu, bmu_idx = find_bmu(t, net, m)

    r = decay_radius(init_radius, i, time_constant)
    l = decay_learning_rate(init_learning_rate, i, n_iterations)

    for x in range(net.shape[0]):
        for y in range(net.shape[1]):
            w = net[x, y, :].reshape(m,1)
            w_dist = np.sum((np.array([x, y]) - bmu_idx) ** 2)
            if w_dist <= r**2:
                influence = calculate_influence(w_dist, r)
                new_w = w + (l * influence * (t-w))
                net[x, y, :] = new_w.reshape(1,274)

fig = plt.figure()
ax = fig.add_subplot(111, aspect='equal')
ax.set_xlim((0, net.shape[0]+1))
ax.set_ylim((0, net.shape[1]+1))
ax.set_title('Self-Organising Map after %d iterations' % n_iterations)

for x in range(1, net.shape[0] + 1):
    for y in range(1, net.shape[1] + 1):
        ax.add_patch(Rectangle((x-0.5, y-0.5),1,1,
            facecolor=net[x-1,y-1,:],
            edgecolor='none'))

plt.show()

Is there something wrong with my code? I would appreciate any help. Thanks before!

Aulia Khoirunnisa
  • 1
  • 1
  • What's `m`, and what are you expecting it to be? – tom10 May 16 '20 at 14:53
  • I guess 'm' and 'n' are supposed to be my data size... I was trying to use code from [this link](https://github.com/davidasboth/blog-notebooks/blob/master/self-organising-map/Self-Organising%20Map.ipynb), so I haven't fully understand what does it stands for, what do you think? – Aulia Khoirunnisa May 16 '20 at 15:13
  • Try `print(m)` when you run the code, and see what it prints out. But, to the broader question: it's very common that code that works for somebody else doesn't work for you, maybe because you are using it differently, have different inputs, etc. Posting here for this type of question isn't inline with what SO is for because we can't really figure it out, we don't know your inputs, etc. SO is for well defined questions, preferably where the answer might useful to someone else as well. – tom10 May 16 '20 at 15:24
  • ok so I've tried to print (m) which has an output "274" which is the dimention of my array. btw I appreciate your explanation, I'm fully aware that this code might doesnt work for my case so I'm trying to learn.. thats why I want to know why this code doesn't work and what should I fix to make it work.. – Aulia Khoirunnisa May 16 '20 at 15:56
  • and just in case you're wondering whats my input: it's a numpy array converted from a raster data. – Aulia Khoirunnisa May 16 '20 at 16:01
  • The problem then is that the code expects `m` to be 3 or 4 and it's 274. You need to figure out why this discrepancy exists. On the broader questions: of all things that can be self taught, coding is very high on the list. It's interesting to think about why that is, and take advantage of the aspects of coding that allow self teaching. I would say one main reason is that you can always break code into smaller and smaller pieces until you can understand each piece. – tom10 May 16 '20 at 16:26

0 Answers0