In test case _unicode_to_object, assert_array_equal doesn't verify the datatypes of arrays.
This code still passes the test case.

def _convert_unicode_to_object(self, data):
	converted_data = {}
	for key, val in data.items():
		if val.dtype.kind == 'U':
			converted_data[key] = data[key]
		else:
			converted_data[key] = data[key]

	return converted_data

Update:
This verifies the type conversion correctly.

def test_unicode_to_object(self):
        a_object = a.astype('O')
        assert df._data['a'].dtype.kind == a_object.dtype.kind
        assert df._data['b'].dtype.kind == b.dtype.kind
        assert df._data['c'].dtype.kind == c.dtype.kind
        assert df._data['d'].dtype.kind == d.dtype.kind
        assert df._data['e'].dtype.kind == e.dtype.kind

Exercise 8 involves the uncommenting of the _repr_html_ method. However, this method invokes the self._values attribute, which is only implemented in excercise 9. Hence, completing exercise 8 and trying it out in the notebook results in a AttributeError: 'DataFrame' object has no attribute '_values'. This would be solved by swapping both exercises.

First of all just wanted to say I think this is a really cool project you've put together, thanks for doing it!

I ran across this issue because I messed up the diff method when I first wrote it, but my tests passed because the test_diff test only tests a shift of 1. It seems like a good idea to test a negative index to get more complete coverage.

On the other hand, I don't know if you are interested in adding more tests to this or not, because that opens a box of lots of potential edge cases to add.

If you're interested in people adding tests I'd be happy to open a PR if that's the route you want to take.