def fuzzy_index_match(possiblities, label, **kwargs):
"""Find the closest matching column label, key, or integer indexed value
Returns:
type(label): sequence of immutable objects corresponding to best matches to each object in label
if label is an int returns the object (value) in the list of possibilities at that index
if label is a str returns the closest str match in possibilities
>>> from collections import OrderedDict as odict
>>> fuzzy_index_match(pd.DataFrame(pd.np.random.randn(9,4), columns=list('ABCD'), index=range(9)), 'b')
'B'
>>> fuzzy_index_match(odict(zip('12345','ABCDE')), 'r2d2')
'2'
>>> fuzzy_index_match(odict(zip('12345','ABCDE')), 1)
'2'
>>> fuzzy_index_match(odict(zip('12345','ABCDE')), -1)
'5'
>>> fuzzy_index_match(odict(zip(range(4),'FOUR')), -4)
0
"""
possibilities = list(possiblities)
if isinstance(label, basestring):
return fuzzy_get(possibilities, label, **kwargs)
if isinstance(label, int):
return possibilities[label]
if isinstance(label, list):
return [fuzzy_get(possibilities, lbl) for lbl in label]
def api(feature='conditions', city='Portland', state='OR', key=None):
"""Use the wunderground API to get current conditions instead of scraping
Please be kind and use your own key (they're FREE!):
http://www.wunderground.com/weather/api/d/login.html
References:
http://www.wunderground.com/weather/api/d/terms.html
Examples:
>>> api('hurric', 'Boise', 'ID') # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
{u'currenthurricane': ...}}}
>>> features = 'alerts astronomy conditions currenthurricane forecast forecast10day geolookup history hourly hourly10day planner rawtide satellite tide webcams yesterday'.split(' ')
>> everything = [api(f, 'Portland') for f in features]
>> js = api('alerts', 'Portland', 'OR')
>> js = api('condit', 'Sacramento', 'CA')
>> js = api('forecast', 'Mobile', 'AL')
>> js = api('10day', 'Fairhope', 'AL')
>> js = api('geo', 'Decatur', 'AL')
>> js = api('hist', 'history', 'AL')
>> js = api('astro')
"""
features = 'alerts astronomy conditions currenthurricane forecast forecast10day geolookup history hourly hourly10day planner rawtide satellite tide webcams yesterday'.split(' ')
feature = util.fuzzy_get(features, feature)
# Please be kind and use your own key (they're FREE!):
# http://www.wunderground.com/weather/api/d/login.html
key = key or env.get('WUNDERGROUND', None, verbosity=-1) or env.get('WUNDERGROUND_KEY', 'c45a86c2fc63f7d0', verbosity=-1)
url = 'http://api.wunderground.com/api/{key}/{feature}/q/{state}/{city}.json'.format(
key=key, feature=feature, state=state, city=city)
return json.load(urllib.urlopen(url))
def make_dataframe(obj, columns=None, exclude=None, limit=1e8):
"""Coerce an iterable, queryset, list or rows, dict of columns, etc into a Pandas DataFrame"""
try:
obj = obj.objects.all()[:limit]
except:
pass
if isinstance(obj, (pd.Series, list, tuple)):
return make_dataframe(pd.DataFrame(obj), columns, exclude, limit)
# if the obj is a named tuple, DataFrame, dict of columns, django QuerySet, sql alchemy query result
# retrieve the "include"d field/column names from its keys/fields/attributes
if columns is None:
columns = get_column_labels(obj)
if exclude is not None and columns is not None and columns and exclude:
columns = [i for i in columns if i not in exclude]
try:
return pd.DataFrame(list(obj.values(*columns)[:limit]))
except:
pass
try:
return pd.DataFrame(obj)[fuzzy_get(obj, columns)]
except:
pass
return pd.DataFrame(obj)
def dataset_from_dataframe(df,
delays=(1, 2, 3),
inputs=(1, 2, -1),
outputs=(-1, ),
normalize=False,
verbosity=1):
"""Compose a pybrain.dataset from a pandas DataFrame
Arguments:
delays (list of int): sample delays to use for the input tapped delay line
Positive and negative values are treated the same as sample counts into the past.
default: [1, 2, 3], in z-transform notation: z^-1 + z^-2 + z^-3
inputs (list of int or list of str): column indices or labels for the inputs
outputs (list of int or list of str): column indices or labels for the outputs
normalize (bool): whether to divide each input to be normally distributed about 0 with std 1
Returns:
3-tuple: tuple(dataset, list of means, list of stds)
means and stds allow normalization of new inputs and denormalization of the outputs
TODO:
Detect categorical variables with low dimensionality and split into separate bits
Vowpel Wabbit hashes strings into an int?
Detect ordinal variables and convert to continuous int sequence
SEE: http://www.stat.berkeley.edu/~breiman/RandomForests/cc_home.htm
"""
if isinstance(delays, int):
if delays:
delays = range(1, delays + 1)
else:
delays = [0]
delays = np.abs(np.array([int(i) for i in delays]))
inputs = [
df.columns[int(inp)] if isinstance(inp, (float, int)) else str(inp)
for inp in inputs
]
outputs = [
df.columns[int(out)] if isinstance(out, (float, int)) else str(out)
for out in (outputs or [])
]
inputs = [fuzzy_get(df.columns, i) for i in inputs]
outputs = [fuzzy_get(df.columns, o) for o in outputs]
N_inp = len(inputs)
N_out = len(outputs)
inp_outs = inputs + outputs
if verbosity > 0:
print("inputs: {}\noutputs: {}\ndelays: {}\n".format(
inputs, outputs, delays))
means, stds = np.zeros(len(inp_outs)), np.ones(len(inp_outs))
if normalize:
means, stds = df[inp_outs].mean(), df[inp_outs].std()
if normalize and verbosity > 0:
print("Input mean values (used to normalize input biases): {}".format(
means[:N_inp]))
print(
"Output mean values (used to normalize output biases): {}".format(
means[N_inp:]))
ds = pb.datasets.SupervisedDataSet(N_inp * len(delays), N_out)
if verbosity > 0:
print(
"Dataset dimensions are {}x{}x{} (records x indim x outdim) for {} delays, {} inputs, {} outputs"
.format(len(df), ds.indim, ds.outdim, len(delays), len(inputs),
len(outputs)))
# FIXME: normalize the whole matrix at once and add it quickly rather than one sample at a time
if delays == np.array([0]) and not normalize:
if verbosity > 0:
print(
"No tapped delay lines (delays) were requested, so using undelayed features for the dataset."
)
assert (df[inputs].values.shape[0] == df[outputs].values.shape[0])
ds.setField('input', df[inputs].values)
ds.setField('target', df[outputs].values)
ds.linkFields(['input', 'target'])
# for inp, outp in zip(df[inputs].values, df[outputs].values):
# ds.appendLinked(inp, outp)
assert (len(ds['input']) == len(ds['target']))
else:
for i, out_vec in enumerate(df[outputs].values):
if verbosity > 0 and i % 100 == 0:
print("{}%".format(i / .01 / len(df)))
elif verbosity > 1:
print('sample[{i}].target={out_vec}'.format(i=i,
out_vec=out_vec))
if i < max(delays):
continue
inp_vec = []
for delay in delays:
inp_vec += list(
(df[inputs].values[i - delay] - means[:N_inp]) /
stds[:N_inp])
ds.addSample(inp_vec, (out_vec - means[N_inp:]) / stds[N_inp:])
if verbosity > 0:
print("Dataset now has {} samples".format(len(ds)))
if normalize:
return ds, means, stds
else:
return ds
def dataset_from_dataframe(df, delays=(1, 2, 3), inputs=(1, 2, -1), outputs=(-1,), normalize=False, verbosity=1):
"""Compose a pybrain.dataset from a pandas DataFrame
Arguments:
delays (list of int): sample delays to use for the input tapped delay line
Positive and negative values are treated the same as sample counts into the past.
default: [1, 2, 3], in z-transform notation: z^-1 + z^-2 + z^-3
inputs (list of int or list of str): column indices or labels for the inputs
outputs (list of int or list of str): column indices or labels for the outputs
normalize (bool): whether to divide each input to be normally distributed about 0 with std 1
Returns:
3-tuple: tuple(dataset, list of means, list of stds)
means and stds allow normalization of new inputs and denormalization of the outputs
TODO:
Detect categorical variables with low dimensionality and split into separate bits
Vowpel Wabbit hashes strings into an int?
Detect ordinal variables and convert to continuous int sequence
SEE: http://www.stat.berkeley.edu/~breiman/RandomForests/cc_home.htm
"""
if isinstance(delays, int):
if delays:
delays = range(1, delays + 1)
else:
delays = [0]
delays = np.abs(np.array([int(i) for i in delays]))
inputs = [df.columns[int(inp)] if isinstance(inp, (float, int)) else str(inp) for inp in inputs]
outputs = [df.columns[int(out)] if isinstance(out, (float, int)) else str(out) for out in (outputs or [])]
inputs = [fuzzy_get(df.columns, i) for i in inputs]
outputs = [fuzzy_get(df.columns, o) for o in outputs]
N_inp = len(inputs)
N_out = len(outputs)
inp_outs = inputs + outputs
if verbosity > 0:
print("inputs: {}\noutputs: {}\ndelays: {}\n".format(inputs, outputs, delays))
means, stds = np.zeros(len(inp_outs)), np.ones(len(inp_outs))
if normalize:
means, stds = df[inp_outs].mean(), df[inp_outs].std()
if normalize and verbosity > 0:
print("Input mean values (used to normalize input biases): {}".format(means[:N_inp]))
print("Output mean values (used to normalize output biases): {}".format(means[N_inp:]))
ds = pb.datasets.SupervisedDataSet(N_inp * len(delays), N_out)
if verbosity > 0:
print("Dataset dimensions are {}x{}x{} (records x indim x outdim) for {} delays, {} inputs, {} outputs".format(
len(df), ds.indim, ds.outdim, len(delays), len(inputs), len(outputs)))
# FIXME: normalize the whole matrix at once and add it quickly rather than one sample at a time
if delays == np.array([0]) and not normalize:
if verbosity > 0:
print("No tapped delay lines (delays) were requested, so using undelayed features for the dataset.")
assert(df[inputs].values.shape[0] == df[outputs].values.shape[0])
ds.setField('input', df[inputs].values)
ds.setField('target', df[outputs].values)
ds.linkFields(['input', 'target'])
# for inp, outp in zip(df[inputs].values, df[outputs].values):
# ds.appendLinked(inp, outp)
assert(len(ds['input']) == len(ds['target']))
else:
for i, out_vec in enumerate(df[outputs].values):
if verbosity > 0 and i % 100 == 0:
print("{}%".format(i / .01 / len(df)))
elif verbosity > 1:
print('sample[{i}].target={out_vec}'.format(i=i, out_vec=out_vec))
if i < max(delays):
continue
inp_vec = []
for delay in delays:
inp_vec += list((df[inputs].values[i - delay] - means[:N_inp]) / stds[:N_inp])
ds.addSample(inp_vec, (out_vec - means[N_inp:]) / stds[N_inp:])
if verbosity > 0:
print("Dataset now has {} samples".format(len(ds)))
if normalize:
return ds, means, stds
else:
return ds
