René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Type promotion error with datetime bins in hist","articleBody":"### Bug report\r\nIt appears that `pyplot.hist` (and I believe the equivalent axis method) does not allow lists of `date` or `datetime` objects to be used as the bin edges, even if the things being binned are dates, so:\r\n\r\n```python\r\nfrom datetime import datetime\r\n\r\nimport matplotlib as mpl\r\nfrom matplotlib import pyplot as plt\r\n\r\ndates = [\r\n    datetime(2019, 1, 5), datetime(2019, 1, 11), datetime(2019, 2, 1),\r\n    datetime(2019, 2, 5), datetime(2019, 2, 18), datetime(2019, 3, 1),\r\n]\r\n\r\ndate_edges = [\r\n    datetime(2019, 1, 1), datetime(2019, 2, 1), datetime(2019, 3, 1),\r\n]\r\n\r\nplt.hist(dates)   # Works\r\nplt.hist(dates, bins=mpl.dates.date2num(date_edges))  # Works\r\nplt.hist(dates, bins=date_edges)   # Fails\r\n```\r\n\r\nThe first `hist` call works but doesn't specify the bins, the second one works and does what we want, but it requires actively converting `date_edges`. The third one *should* do the same thing as the second one, but instead it fails with:\r\n\r\n\u003e TypeError: '\u003c' not supported between instances of 'float' and 'datetime.datetime'\r\n\r\nFull error message below:\r\n\r\n\u003cdetails\u003e\r\n\r\n```\r\nTypeError                                 Traceback (most recent call last)\r\n\u003cipython-input-27-5b07a69daccb\u003e in \u003cmodule\u003e\r\n----\u003e 1 plt.hist(dates, bins=date_bins)\r\n\r\n/tmp/mpl/lib/python3.7/site-packages/matplotlib/pyplot.py in hist(x, bins, range, density, weights, cumulative, bottom, histtype, align, orientation, rwidth, log, color, label, stacked, normed, data, **kwargs)\r\n   2640         align=align, orientation=orientation, rwidth=rwidth, log=log,\r\n   2641         color=color, label=label, stacked=stacked, normed=normed,\r\n-\u003e 2642         **({\"data\": data} if data is not None else {}), **kwargs)\r\n   2643\r\n   2644\r\n\r\n/tmp/mpl/lib/python3.7/site-packages/matplotlib/__init__.py in inner(ax, data, *args, **kwargs)\r\n   1599     def inner(ax, *args, data=None, **kwargs):\r\n   1600         if data is None:\r\n-\u003e 1601             return func(ax, *map(sanitize_sequence, args), **kwargs)\r\n   1602\r\n   1603         bound = new_sig.bind(ax, *args, **kwargs)\r\n\r\n/tmp/mpl/lib/python3.7/site-packages/matplotlib/axes/_axes.py in hist(self, x, bins, range, density, weights, cumulative, bottom, histtype, align, orientation, rwidth, log, color, label, stacked, normed, **kwargs)\r\n   6765             # this will automatically overwrite bins,\r\n   6766             # so that each histogram uses the same bins\r\n-\u003e 6767             m, bins = np.histogram(x[i], bins, weights=w[i], **hist_kwargs)\r\n   6768             m = m.astype(float)  # causes problems later if it's an int\r\n   6769             if mlast is None:\r\n\r\n\u003c__array_function__ internals\u003e in histogram(*args, **kwargs)\r\n\r\n/tmp/mpl/lib/python3.7/site-packages/numpy/lib/histograms.py in histogram(a, bins, range, normed, weights, density)\r\n    869             for i in _range(0, len(a), BLOCK):\r\n    870                 sa = np.sort(a[i:i+BLOCK])\r\n--\u003e 871                 cum_n += _search_sorted_inclusive(sa, bin_edges)\r\n    872         else:\r\n    873             zero = np.zeros(1, dtype=ntype)\r\n\r\n/tmp/mpl/lib/python3.7/site-packages/numpy/lib/histograms.py in _search_sorted_inclusive(a, v)\r\n    452     \"\"\"\r\n    453     return np.concatenate((\r\n--\u003e 454         a.searchsorted(v[:-1], 'left'),\r\n    455         a.searchsorted(v[-1:], 'right')\r\n    456     ))\r\n\r\nTypeError: '\u003c' not supported between instances of 'float' and 'datetime.datetime'\r\n```\r\n\u003c/details\u003e\r\n\r\nI think this is another manifestation of the same underlying problem we saw in #12863 (see my comments in there for some more details on that), which is that a lot of the `datetime` stuff converts somewhat eagerly to floats but *some* of it doesn't.\r\n\r\nI think this case is easier than #12863 in the sense that you can almost certainly translate this into an unambiguous point on the number line and eagerly convert the bin edges early on, depending on how `matplotlib` treats naive datetimes. It's possible that you may want to detect if `x` and `bins` are both naive or both aware and throw an error if they aren't, but I don't see that as a major sticking point.\r\n\r\n**Matplotlib version**\r\n  * Operating system: Linux\r\n  * Matplotlib version: 3.1.1\r\n  * Matplotlib backend (`print(matplotlib.get_backend())`): TkAgg\r\n  * Python version: 3.7.4","author":{"url":"https://github.com/pganssle","@type":"Person","name":"pganssle"},"datePublished":"2019-09-23T19:50:21.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":4},"url":"https://github.com/15332/matplotlib/issues/15332"}