# Which columns are most strongly related The strongest relationships are between the descriptive and classifier columns (sub-vertical, industry vertical, remarks, investment type, city). The amount column is only weakly tied to everything else. I measured association with mutual information (MI, in nats) and normalized mutual information (NMI, 0 to 1). I used these instead of Pearson correlation because almost every column is categorical. Amount was bucketed on a log10 scale and year was taken from the date. The ranked pairs are in the step-1 result above. **Strongest pairs (NMI)** - **Sub-vertical and year (0.974):** this looks nearly deterministic, but it is largely an artifact. Sub-vertical has 1,943 distinct values across 3,044 rows, so most values appear only once or twice and trivially "identify" a year. - **Industry vertical and remarks (0.888, MI 1.07):** remarks has only 75 distinct values and is closely tied to the vertical, so it probably carries sector-specific notes or tags. - **Investment type and sub-vertical (0.769), city and sub-vertical (0.694), and amount bucket and sub-vertical (0.655):** these are inflated by the same high-cardinality effect. - **Industry vertical and sub-vertical (NMI 0.525, MI 2.02):** this is the highest MI of any pair and is the most credible relationship. A sub-vertical is a refinement of its parent vertical. - **Industry vertical and year (0.582), and remarks and sub-vertical (0.526):** moderate links. **Moderate, more trustworthy links** (the lower-cardinality columns) - City and industry vertical: NMI 0.425, MI 1.10. Sectors cluster in certain cities. - Industry vertical and investment type: NMI 0.419. - Amount bucket and industry vertical: NMI 0.299. - Amount bucket and investment type: NMI 0.267. This is the expected link between round type and ticket size, but it is modest. **Weak or negligible** - Amount bucket and year: 0.036. - Amount bucket and city: 0.070. - Amount bucket and remarks: 0.112. - Investment type and remarks: 0.176. - City and remarks: 0.173. **Caveats** - NMI rises with the number of distinct values, so pairs involving sub-vertical (1,943 distinct) and industry vertical (822 distinct) are overstated. The step-3 profile shows these are messy free-text-like columns. - Only 2,066 of 3,044 amounts parse as numbers (the median parsed amount is $1.725M), so any amount-related association rests on about two-thirds of the rows. - Startup name and investors name are unique per row, so they have no meaningful association with anything. Sr No is just an identifier. In short, the vertical hierarchy (industry vertical, sub-vertical, remarks) is the tightest cluster. City and investment type relate moderately to it, and funding amount is only loosely connected to any other attribute.