def pca_signatures(variance, return_pca=False):
"""
Does a principal component analysis on the signature dataframe so that the
number of principal components accounts for at least 80 % of the total
variance of the original variables
:param variance: amount of variance that is to be explained
:param return_pca: allows returning of the pca object
"""
# Get the data
meta_df = read_attributes_signatures.read_meta()
att_df, sig_df = read_attributes_signatures.seperate_attributes_signatures(
meta_df)
# Perform the pca
pca = PCA(n_components=variance, svd_solver="full")
# Standardize the data, so the PCA makes more sense
standardized_df = StandardScaler().fit_transform(sig_df)
# Calculate the components
principal_components = pca.fit_transform(np.array(standardized_df))
print("Explained variance of the components (sorted in ascending order):")
print(pca.explained_variance_ratio_)
# Make it a dataframe again
principal_df = pd.DataFrame(data=principal_components, index=sig_df.index)
# Give the columns more meaningful names
principal_df.columns = [
"PC " + str(i) for i in range(1,
len(principal_df.columns) + 1)
]
if return_pca:
return pca
else:
return principal_df
def read_data():
meta_df = read_attributes_signatures.read_meta()
att_df, sig_df = read_attributes_signatures.seperate_attributes_signatures(
meta_df)
knoben = pd.read_csv("catchment_clusters_with_continoues_climate.csv",
index_col=1)
df = pd.merge(sig_df, knoben, right_index=True, left_index=True)
df.drop(["FID", "gauge_lat", "gauge_lon", "b1_", "b2_", "b3_"],
axis=1,
inplace=True)
df.columns = [
'Mean annual discharge', 'Mean winter discharge',
'Mean half-flow date', 'Q95 (high flow)', 'Runoff ratio',
'Mean summer discharge', "Catchment Cluster", "Aridity", "Seasonality",
"Precipitation falling as snow"
]
# Rescale the data to the original values of Wouter
# Find the new min max values by looking how far the current min max values
# are away to the min max of the range.
df["Aridity"] = rescaler(df["Aridity"], 1, 0, 1, -1)
df["Seasonality"] = rescaler(df["Seasonality"], 1, 0, 2, 0)
return df
frameon=True, fancybox=True)
legend.get_frame().set_edgecolor("grey")
legend.get_frame().set_facecolor("white")
# Make plot nicer by removing the borders
ax.set_facecolor("white")
for spine in ax.spines.values():
spine.set_visible(False)
# Add correct descriptions
ax.set_title(describer, alpha=alpha)
ax.set_ylabel("PC 2", alpha=alpha)
ax.set_xlabel("PC 1", alpha=alpha)
ax.grid(color="grey", alpha=alpha)
plt.setp(ax.get_yticklabels(), alpha=alpha)
plt.setp(ax.get_xticklabels(), alpha=alpha)
ax.tick_params(axis=u'both', which=u'both',length=0)
# Save the plot
fig.tight_layout()
plt.savefig(describer.replace("\n","") + ".png", bbox_inches="tight")
plt.close()
if __name__ == "__main__":
variance = 0.8
pca_df = pca.pca_signatures(variance)
meta_df = read_attributes_signatures.read_meta()
att_df, sig_df = read_attributes_signatures.seperate_attributes_signatures(meta_df)
plotting_df = pd.concat([pca_df, att_df], axis=1)
plot_pca(plotting_df, att_df.columns)