Cannot append DataFrames with uint dtypes to HDFStore

[jmellen]

Trying to store DataFrames with unsigned integer dtypes in HDFStore fails due to a bug in the get_atom_data methods of the DataCol and DataIndexableCol classes in pandas.io.pytables. These methods use Python's capitalize method to map dtypes to PyTables column types, and fail because PyTables' unsigned int classes start with two capital letters (e.g., UInt32Col). This error occurs on 0.11 and the current master:

import pandas as pd
import numpy as np

uint8_series = pd.Series(np.random.random_integers(0,high=255,size=5), dtype=np.uint8)
udf = pd.DataFrame({'u08': uint8_series}, index=np.arange(5))

store = pd.HDFStore('uint.h5')

# this invocation will throw an error in pandas 0.11 and current master
store.append('uints', udf)

I have a commit + test on my fork that fixes this, just needed to submit the bug here first to have the right commit message on my fork, per contribution standards. First time committer.

[jreback]

good catch. I should have had a mapping table rather than do capitalize for just this reason....(after all there are not that many types)....and could provide a more informative message as well....put up your PR and we'll take a look (also enable travis!, see CONTRIBUTION in the main dir for how)

[jmellen]

Travis is running right now; the fix just checks for startswith('uint') but a mapping table would also work. I didn't have enough confidence in what values self.kind could have (is it guaranteed to be lowercase?) to make a mapping.

Will send the PR when Travis finishes.

[jreback]

your PR is good (just need a release notes mention)....if it can't find the dtype it'll raise anyhow (as you discovered)

also..if you are interested, pls check out #2391

[jreback]

closed by #3493