main.py

import numpy as np
from sompy.sompy import SOMFactory
import pandas as pd
import glob
import os


# read in all csvs from folder
path = '..\\..\\data\\'
all_files = glob.glob(os.path.join(path, "*.csv"))

# concat into one df
df_from_each_file = (pd.read_csv(f, skiprows = 31) for f in all_files)
concatenated_df   = pd.concat(df_from_each_file, ignore_index=True)

# get columns Lat, Long, Mean Temp, Max Temp, Min temp, Precipitation
data = concatenated_df[['Lat', 'Long', 'Tm', 'Tx', 'Tn', 'P']]
data = data.apply(pd.to_numeric,  errors='coerce')
data = data.dropna(how='any')
names = ['Latitude', "longitude", 'Monthly Median temperature (C)','Monthly Max temperature (C)', 'Monthly Min temperature (C)', 'Monthly total precipitation (mm)']

print(data.head())

# create the SOM network and train it. You can experiment with different normalizations and initializations
sm = SOMFactory().build(data.values, normalization = 'var', initialization='pca', component_names=names)
sm.train(n_job=1, verbose=False, train_rough_len=2, train_finetune_len=5)

# The quantization error: average distance between each data vector and its BMU.
# The topographic error: the proportion of all data vectors for which first and second BMUs are not adjacent units.
topographic_error = sm.calculate_topographic_error()
quantization_error = np.mean(sm._bmu[1])
print ("Topographic error = %s; Quantization error = %s" % (topographic_error, quantization_error))

# component planes view
from sompy.visualization.mapview import View2D
view2D  = View2D(10,10,"rand data",text_size=12)
view2D.show(sm, col_sz=4, which_dim="all", desnormalize=True)

# U-matrix plot
from sompy.visualization.umatrix import UMatrixView

umat  = UMatrixView(width=10,height=10,title='U-matrix')
umat.show(sm)

# do the K-means clustering on the SOM grid, sweep across k = 2 to 20
from sompy.visualization.hitmap import HitMapView
K = 20 # stop at this k
sm.cluster(K)
hits  = HitMapView(20,20,"Clustering",text_size=12)
a=hits.show(sm)