Data Visualization Guide for Multi-dimensional Data

Inference: Now we can see that in some of the columns, there are a few Null values like Skin thickness, Insulin, BMI, etc.

Viewing the Data

So, surprisingly no one it’s useful to view the data. Straight up by using head, we can see that this dataset is utilizing 0 to represent no value – unless some poor unfortunate soul has a skin thickness of 0.

If we want to do more than expect the data, we can use the described function we talked about in the previous section.

df.describe()

Output:

Scatter Matrix

Scatter Matrix is one of the best plots to determine the relationship (linear mostly) between the multiple variables of all the datasets; when you will see the linear graph between two or more variables that indicates the high correlation between those features, it can be either positive or negative correlation.

pd.plotting.scatter_matrix(df, figsize=(10, 10));

Output:

Inference: From the above plot, we can say that this plot alone is quite descriptive as it is showing us the linear relationship between all the variables in the dataset. For example, we can see that skin Thickness and BMI is sharing linear tendencies.

Note: Due to the big names of columns, we are facing a bit issue while reading the same though that can be improved (out of the scope of the article).

df2 = df.dropna()
colors = df2["Outcome"].map(lambda x: "#44d9ff" if x else "#f95b4a")
pd.plotting.scatter_matrix(df2, figsize=(10,10), color=colors);

Output:

Inference: The scatter plot gives us both the histograms for the distributions along the diagonal and also a lot of 2D scatter plots off-diagonal. Not that this is a symmetric matrix, so I just look at the diagonal and below it normally. We can see that some variables have a lot of scattering, and some are correlated (ie, there is a direction in their scatter). This leads us to another type of plot i.e., correlation plot.

Correlation Plots

Before going into a deep discussion with the correlation plot, we first need to understand the correlation and for that reason, we are using the pandas’ corr() method that will return the Pearson’s correlation coefficient between two data inputs. In a nutshell, these plots easily quantify which variables or attributes are correlated with each other.

df.corr()

Output:

sb.set(rc={'figure.figsize':(11,6)})
sb.heatmap(df.corr());

Output:

Inference: In a seaborn or matplotlib supported correlation plot, we can compare the higher and lower correlation between the variables using its color palette and scale. In the above graph, the lighter the color, the more the correlation and vice versa. There are some drawbacks in this plot which we will get rid of in the very next graph.

sb.heatmap(df.corr(), annot=True, cmap="viridis", fmt="0.2f");

Output:

Inference: Now one can see this is a symmetric matrix too. But it immediately allows us to point out the most correlated and anti-correlated attributes. Some might just be common sense – Pregnancies v Age for example – but some might give us a real insight into the data.

Here, we have also used some parameters like annot= True so that we can see correlated values and some formatting as well.

2D Histograms

2D Histograms are mainly used for image processing, showing the intensities of pixels at a certain position of the image. Similarly, we can also use it for other problem statements, where we need to analyze two or more variables as two-dimensional or three-dimensional histograms, which provide multi-dimensional Data.

For the rest of this section, we’re going to use a different dataset with more data.

Note: 2-D histograms are very useful when you have a lot of data. See here for the API.

df2 = pd.read_csv("height_weight.csv")
df2.info()
df2.describe()

Output:

plt.hist2d(df2["height"], df2["weight"], bins=20, cmap="magma")
plt.xlabel("Height")
plt.ylabel("Weight");

Output:

hist, x_edge, y_edge = np.histogram2d(df2["height"], df2["weight"], bins=20)
x_center = 0.5 * (x_edge[1:] + x_edge[:-1])
y_center = 0.5 * (y_edge[1:] + y_edge[:-1])

plt.contour(x_center, y_center, hist, levels=4)
plt.xlabel("Height")
plt.ylabel("Weight");