Unless you are part of a vanishingly small group of autodidacts who can train themselves up to graduate school levels of expertise in multiple overlapping subjects - statistics, computer science (might be able to get away with just being an ok programmer), and the interdisciplinary combination of those called "machine learning," you should disappear into a statistics degree program, and amend the traditional stats program deficiencies with the modern-day leavening agents that create "machine learning."

Downloading scikitlearn and R and such is not going to work. At that level you are only qualified to be bossed around by a real scientist or statistician. You are an "analyst".