def get_combined_frame(input_frame, result_frame, include_on_out_fields):
combined_frame = dt.Frame()
for include_column in include_on_out_fields:
combined_frame = dt.cbind(combined_frame, input_frame[include_column])
combined_frame = dt.cbind(combined_frame, result_frame)
return combined_frame
def test_cbind_issue2024():
DT = dt.Frame([[]] * 2, names=["A.1", "A.5"])
with pytest.warns(DatatableWarning):
RZ = dt.cbind(DT, DT)
assert RZ.names == ("A.1", "A.5", "A.2", "A.6")
RZ = dt.cbind(DT, DT, DT)
assert RZ.names == ("A.1", "A.5", "A.2", "A.6", "A.3", "A.7")
RZ = dt.cbind(DT, DT, DT, DT)
assert RZ.names == ("A.1", "A.5", "A.2", "A.6", "A.3", "A.7", "A.4",
"A.8")
RZ = dt.cbind(DT, DT, DT, DT, DT)
assert RZ.names == ("A.1", "A.5", "A.2", "A.6", "A.3", "A.7", "A.4",
"A.8", "A.9", "A.10")
def test_issue1905():
# cbind() is passed a generator, where each generated Frame is a
# temporary. In this case cbind() should take care to keep the
# references to all frames while working, lest they get gc-d by
# the end of the generator loop.
DT = dt.cbind(dt.Frame(range(50), names=[str(i)]) for i in range(30))
assert DT.shape == (50, 30)
def test_issue2055(numpy):
DT = dt.cbind(dt.Frame(A=[1, 2]),
dt.Frame(numpy.ma.array([True, True], mask=[False, False])))
DT.nrows = 1
DT = DT.copy()
frame_integrity_check(DT)
assert DT.to_list() == [[1], [True]]
def test_issue1921():
n = 1921
DTA = dt.Frame(A=range(n))
DTB = dt.repeat(dt.Frame(B=["hey"], stype=dt.str64), n)
DT = dt.cbind(DTA, DTB)
out = DT.to_csv()
assert out == "\n".join(["A,B"] + ["%d,hey" % i for i in range(n)] + [""])
def test_columnset_sum(DT):
assert_equals(DT[:, f[int].extend(f[float])], DT[:, [int, float]])
assert_equals(DT[:, f[:3].extend(f[-3:])], DT[:, [0, 1, 2, -3, -2, -1]])
assert_equals( DT[:, f['A','B','C'].extend(f['E','F', 'G'])], DT[:, [0, 1, 2, -3, -2, -1]])
assert_equals(DT[:, f.A.extend(f.B)], DT[:, ['A', 'B']])
assert_equals(DT[:, f[:].extend({"extra": f.A + f.C})],
dt.cbind(DT, DT[:, {"extra": f.A + f.C}]))
def py_tidy_descriptive_stats(DT):
"""Generate summary statistics of datatable"""
datos_dict = DT.to_dict()
summary_stats_of_dict = {
k: [
np.nanmean(v),
np.nanmedian(v),
np.nanmin(v),
np.nanmax(v),
np.nanstd(v),
np.percentile(v, 25, interpolation='midpoint'),
np.percentile(v, 75, interpolation='midpoint'),
np.percentile(v, 75, interpolation='midpoint') -
np.percentile(v, 25, interpolation='midpoint'),
np.nanstd(v) / np.sqrt(np.shape(v)[0])
]
for k, v in datos_dict.items()
}
summary_dict_names = dt.Frame({
'descriptive_stats':
['Mean', 'Median', 'Min', 'Max', 'Std', 'Q1', 'Q3', 'IQR', 'SE']
})
summary_stats_of_dict_prep = {
k: list(map(lambda x: np.round(x, 3), v))
for k, v in summary_stats_of_dict.items()
}
summary_stat_dt = dt.Frame(summary_stats_of_dict_prep)
return dt.cbind(summary_dict_names, summary_stat_dt)
def test_materialize():
DT1 = dt.Frame(A=range(12))[::2, :]
DT2 = dt.repeat(dt.Frame(B=["red", "green", "blue"]), 2)
DT3 = dt.Frame(C=[4, 2, 9.1, 12, 0])
DT = dt.cbind(DT1, DT2, DT3, force=True)
assert frame_columns_virtual(DT) == (True, True, True)
DT.materialize()
assert frame_columns_virtual(DT) == (False, False, False)
def test_aggregate_3d_fixed_small_radius():
DT = dt.Frame([range(10)] * 3)
[DTE, DTM] = aggregate(DT, min_rows=0, nd_max_bins=1, fixed_radius=0.1)
DTE_ref = cbind(
DT, dt.Frame([1] * 10 / dt.stype.int32, names=["members_count"]))
DTM_ref = dt.Frame(range(10), names=["exemplar_id"])
assert_equals(DTE, DTE_ref)
assert_equals(DTM, DTM_ref)
def test_cbind_method():
d0 = dt.Frame({"A": [1, 2, 3]})
d1 = dt.Frame({"B": list('abc')})
d2 = dt.Frame({"C": [5.6, 7.1, -3.3]})
dr = dt.cbind(d0, d1, d2)
assert dr.names == ("A", "B", "C")
res = dt.Frame([[1, 2, 3], ["a", "b", "c"], [5.6, 7.1, -3.3]],
names=("A", "B", "C"))
assert_equals(dr, res)
def test_cbind_notinplace():
d0 = dt.Frame({"A": [1, 2, 3]})
d1 = dt.Frame({"B": [4, 5, 6]})
dt_compute_stats(d0, d1)
dd = dt.cbind(d0, d1)
dr = dt.Frame({"A": [1, 2, 3], "B": [4, 5, 6]})
assert_equals(dd, dr)
assert_equals(d0, dt.Frame({"A": [1, 2, 3]}))
assert_equals(d1, dt.Frame({"B": [4, 5, 6]}))
def test_debug_logger_default_with_report_args(capsys):
assert dt.options.debug.logger is None
with dt.options.debug.context(enabled=True, report_args=True):
assert dt.options.debug.logger is None
assert dt.options.debug.enabled is True
DT = dt.Frame(range(100000))
out, err = capsys.readouterr()
print(out)
assert not err
assert re.search(r"<Frame#[\da-fA-F]+>.__init__\(range\(0, 100000\)\)", out)
assert re.search(r"# \d+(?:\.\d+)?(?:[eE][+-]?\d+)? s", out)
with pytest.raises(TypeError):
dt.cbind(3)
out, err = capsys.readouterr()
assert not err
assert "datatable.cbind(3) {" in out
assert re.search(r"} # \d+(?:\.\d+)?(?:[eE][+-]?\d+)? s \(failed\)", out)
def test_materialize():
DT1 = dt.Frame(A=range(12))[::2, :]
DT2 = dt.repeat(dt.Frame(B=["red", "green", "blue"]), 2)
DT3 = dt.Frame(C=[4, 2, 9.1, 12, 0])
DT = dt.cbind(DT1, DT2, DT3, force=True)
assert frame_column_rowindex(DT, 0).type == "slice"
assert frame_column_rowindex(DT, 1).type == "arr32"
assert frame_column_rowindex(DT, 2) is None
DT.materialize()
assert frame_column_rowindex(DT, 0) is None
assert frame_column_rowindex(DT, 1) is None
assert frame_column_rowindex(DT, 2) is None
def test_topandas_view_mixed():
d0 = dt.Frame(A=range(100))
d1 = d0[7:17, :]
d2 = dt.Frame(B=['foo', 'bar', 'buzz'] * 3 + ['finale'])
d3 = dt.Frame(V=[2.2222])
d3.nrows = 10
dd = dt.cbind(d1, d2, d3)
pp = dd.to_pandas()
assert pp.columns.tolist() == ["A", "B", "V"]
assert pp["A"].tolist() == list(range(7, 17))
assert pp["B"].tolist() == d2.to_list()[0]
assert pp["V"].tolist()[0] == 2.2222
assert all(math.isnan(x) for x in pp["V"].tolist()[1:])
def pd_dt_concat(frames, axis=0):
"""
Concatenate sequence of datatable Frames or pandas DataFrames `frames` along `axis` (0 means rows, 1 means columns).
"""
if USE_DT:
if axis == 0:
return dt.rbind(*frames)
elif axis == 1:
return dt.cbind(*frames)
else:
raise ValueError('invalid axis:', axis)
else:
return pd.concat(frames, axis=axis)
def write_table(df, name, output_dir, add_index=True):
"""
Add a primary key to df ('id' column) and write it to output_dir
as a .csv file.
@param df: [`datatable.Frame`] A PharmacoDB table
@param name: [`string`] The name of the table
@param output_dir: [`string`] The directory to write the table to
@return: [`datatable.Frame`] The indexed PharmacoDB table
"""
print(f'Writing {name} table to {output_dir}...')
if add_index:
# Index datatable
df = cbind(dt.Frame(id=np.arange(df.nrows) + 1), df)
# Write to .csv
df.to_csv(os.path.join(output_dir, f'{name}.csv'))
return df
def pydt_reshape_wide_to_long(DT,*measure_vars,var_name=None,val_name=None):
"""reshaping datatable from wide columns to long """
dt_cols=[*measure_vars]
measure_col_dict = DT[:,[*measure_vars]].to_dict()
variables_dict={'variable':[],'value':[]}
for k,v in measure_col_dict.items():
variables_dict['variable'].extend(repeat(k,len(v)))
variables_dict['value'].extend(v)
wide_to_long_dt = dt.Frame(variables_dict)
removed_cols_dt = DT[:,f[:].remove([ f[col] for col in dt_cols])].to_dict()
non_measures_dt = dt.Frame({k:list(chain.from_iterable(list(repeat(v,len(dt_cols))))) for k,v in removed_cols_dt.items()})
if var_name and val_name is not None:
wide_to_long_dt.names={'variable':var_name,'value':val_name}
wide_to_long_prep_dt=dt.cbind(non_measures_dt,wide_to_long_dt)
return wide_to_long_prep_dt
def test_cbind_api():
DT1 = dt.Frame(A=[1, 2, 3])
DT2 = dt.Frame(B=[-4, -5, None])
DT3 = dt.Frame(X=["makes", "gonna", "make"])
RES1 = dt.cbind(DT1, DT2, DT3)
RES2 = dt.cbind([DT1, DT2, DT3])
RES3 = dt.cbind((DT1, DT2, DT3)) # tuple
RES4 = dt.cbind([DT1], [DT2, DT3])
RES5 = dt.cbind(DT1, [DT2], DT3)
RES6 = dt.cbind((frame for frame in [DT1, DT2, DT3])) # generator
assert_equals(RES1, RES2)
assert_equals(RES1, RES3)
assert_equals(RES1, RES4)
assert_equals(RES1, RES5)
assert_equals(RES1, RES6)
def dtm_to_datatable(dtm, doc_labels, vocab, colname_rowindex='_doc'):
"""
Convert a (sparse) DTM to a datatable Frame using document labels `doc_labels` as row idenitifier (with column name
`colname_rowindex`) and `vocab` as column names.
.. seealso:: :func:`~tmtoolkit.bow.dtm.dtm_to_dataframe` for generating a pandas DataFrame.
:param dtm: (sparse) document-term-matrix of size NxM (N docs, M is vocab size) with raw terms counts
:param doc_labels: document labels used as row index (row names); size must equal number of rows in `dtm`
:param vocab: list or array of vocabulary used as column names; size must equal number of columns in `dtm`
:param colname_rowindex: column name for row identifier (i.e. column where the document labels are put)
:return: datatable Frame
"""
if not USE_DT:
raise RuntimeError('this function requires the package "datatable" to be installed')
import datatable as dt
if dtm.ndim != 2:
raise ValueError('`dtm` must be a 2D array/matrix')
if dtm.shape[0] != len(doc_labels):
raise ValueError('number of rows must be equal to `len(doc_labels)')
if dtm.shape[1] != len(vocab):
raise ValueError('number of rows must be equal to `len(vocab)')
if isinstance(dtm, np.matrix):
dtm = dtm.A
if not isinstance(dtm, np.ndarray):
dtm = dtm.toarray()
return dt.cbind(dt.Frame({colname_rowindex: doc_labels}),
dt.Frame(dtm, names=list(vocab)))
def test_cbind_nones():
DT = dt.cbind(None, dt.Frame(A=range(5)), None, dt.Frame(B=[0] * 5))
assert_equals(DT, dt.Frame(A=range(5), B=[0] * 5))
}) ##### Notes: * Now have a pandas DF **seattle_dates_df** with these columns- year,month,day,hour and week_day, and it needs to be converted to a DT for further analysis * Here, a DF should be passed to **dt.Frame()** method and assigned to a new DT with name as illustrated below # Convering date df to dt seatle_dates_dt = dt.Frame(seatle_dates_df) * In our original seattle DT we no more require date column in string format, so better to remove it from DT # delating a string date column from dt del seattle_bikes_dt['date'] * The DT's seattle_dates_dt and seattle_bikes_dt will be concatenated with the help of **dt.cbind()** function # Concatinating two dts to have a tidy dt seatle_bikes_dt_tidy = dt.cbind(seatle_dates_dt,seattle_bikes_dt) # Tidy DT first 4 and last 4 observations seatle_bikes_dt_tidy ##### Notes: * We can see that a column crossing has a different categorical level values, and 3 of them are having longer names, to make them shorter their values should be updated/modified with shorter string values. * In Datatable it can be achived as done in below 4 code chunks. * here is a simple syntax for this: DT[column value condition, the column to be updated]='new value' # Modifying observations of crossing col - set - 1 seatle_bikes_dt_tidy[f.crossing=="39th Ave NE Greenway at NE 62nd St",f.crossing]='Greenwayway-NE-62Strt'
def join(names1, names2):
with pytest.warns(DatatableWarning):
return dt.cbind(dt.Frame(names=names1),
dt.Frame(names=names2)).names
def test_cbind_0rows_3():
DT0 = dt.Frame(A=[], B=[], C=[])
RES1 = dt.cbind(dt.Frame(), DT0)
RES2 = dt.cbind(DT0, dt.Frame())
assert_equals(RES1, DT0)
assert_equals(RES2, DT0)
def test_cbind_0rows_1():
"""Issue #1604."""
res = dt.cbind(dt.Frame(A=[]), dt.Frame(B=[]))
assert res.names == ("A", "B")
assert res.shape == (0, 2)
def create_data(X: dt.Frame = None):
""" Convert transactional data to i.i.d. data by making time-based aggregations """
if X is None:
X = TransactionalToIID.make_transactions()
if not make_features_from_scratch:
return {'raw_transactions_non_iid': X}
if X is None:
return []
X_pd = X[:, [col_date, col_group, target]].to_pandas() # faster, since only working on a few cols
X_pd[col_row_id] = np.arange(X_pd.shape[0])
y = X_pd[target]
y_enc = target + ".enc"
# Create boolean target
X_pd[y_enc] = (y == target_labels[1]).astype(int)
# Make sure time is datetime64, not string
X_pd[col_date] = pd.to_datetime(X_pd[col_date])
for leak in leaky_choices:
# Create the groups
groups = X_pd.groupby(col_group)
shift_amount = 0 if leak else 1 # this is critical to avoid leaks! DO NOT SET IT TO 0 IN PRODUCTION!
# Compute aggregation over time
for t in window_length_days:
t_days = str(t) + "d" # pandas will do rolling window over this many days ('5d' etc.)
for op in operators:
lag_feature = []
for _, df in groups:
df = df.sort_values(col_date)
time_window = df.set_index(col_date)[y_enc].shift(shift_amount). \
dropna().rolling(t_days, min_periods=1) # get time window. if leaky, includes self
res = getattr(time_window, op)() # apply operator on time window
res.index = df.index[shift_amount:]
lag_feature.append(res)
# Index is set on both side so equal works and reorders rows automatically
X_pd["%s%s_%s_past_%d_days_grouped_by_%s" %
("leaky_" if leak else "", op, target, t, col_group)] = pd.concat(lag_feature, axis=0)
del X_pd[y_enc] # delete temporary binary response column
# delete grouping column, since have all aggregations already in iid form
del X_pd[col_group]
del X[col_group]
# create datatable frame of new features (only)
X_features = dt.Frame(X_pd.loc[:, [x for x in X_pd.columns if x not in [col_date, target, col_row_id]]])
# add new features to original frame
X_new = dt.cbind(X, X_features)
out = {}
for name, time_range in {
# 2-way split: ideal for iid, let Driverless do internal validation splits on training split
'train_iid': X_pd[col_date] <= split_date,
'test_iid': X_pd[col_date] > split_date
}.items():
# X_pd is pandas - easier to deal with time slices, and keep row_id to index into datatable below
which_rows = X_pd.loc[time_range, col_row_id].reset_index(drop=True).values
if shuffle:
np.random.shuffle(which_rows) # shuffle data for generality - no impact on iid modeling
name += ".shuf"
for leak in leaky_choices:
X_out = X_new.copy() # shallow copy
if leak:
cols_to_del = [x for x in X_features.names if "leaky" != x[:5]]
else:
cols_to_del = [x for x in X_features.names if "leaky" == x[:5]]
del X_out[:, cols_to_del]
out[name + (".leaky" if leak else "")] = X_out[which_rows, :]
return out
py_dt_two_group_proportions_summary(policia_tidy_dt,'driver_race','is_arrested')
# stop time
stop_time_df = policia_tidy_dt[:,(f.stop_time)].to_pandas()
# extracting hour
stop_time_hour = stop_time_df.stop_time.str.extract(r'([\d]{2})')
# a new dataframe
stop_time_hour_dt = dt.Frame(stop_time_hour)
# change a col name
stop_time_hour_dt.names={'0':'stop_hour'}
# Binding two dts
policia_tidy_dt_v1 = dt.cbind(policia_tidy_dt,stop_time_hour_dt)
# Hour wise arrests
hour_wise_arrests_dt = py_dt_two_group_proportions_summary(policia_tidy_dt_v1,'stop_hour','is_arrested')
# Visualization
alt.Chart(hour_wise_arrests_dt.to_pandas()).mark_bar().encode(
alt.X('stop_hour:N'),
alt.Y('count'),
alt.Color('is_arrested')
).properties(
title= 'Hour wise arrest trends'
)
# Hour wise arrest rates
amigos_info_dt[:,dt.update(temp=f.directed_by==f.written_by)]
# are the directors and writers same for a title ?
amigos_info_dt[f.temp==True,:]
# remove the temp col
del amigos_info_dt["temp"]
# split writers column
writers_list = [ elemento.split('&') for elemento in amigos_info_dt[:,f.written_by].to_list()[0] ]
# create a new DT with writers
writers_dt = dt.Frame({'no_of_writers':[len(elem) for elem in writers_list]})
# Joining two DTs
amigos_info_dt_v1 = dt.cbind(amigos_info_dt,writers_dt)
# No of writers
alt.Chart(amigos_info_dt_v1[:,count(),by(f.no_of_writers)].to_pandas()).mark_bar().encode(
alt.X('count'),
alt.Y('no_of_writers:O')
).properties(
title='Number of writers in titles'
)
amigos_year = dt.Frame({'year':[re.findall(r'[\d]{4}',fecha)[0] for fecha in amigos_info_dt_v1[:,f.air_date].to_list()[0] ]})
alt.Chart(amigos_year[:,count(),by(f.year)].to_pandas()).mark_line().encode(alt.X('year'),alt.Y('count'))
def test_cbind_empty3():
DT = dt.cbind()
assert_equals(DT, dt.Frame())
def test_create_from_doublestar_expansion():
DT0 = dt.Frame(A=range(3), B=["df", "qe;r", None])
DT1 = dt.Frame(D=[7.99, -12.5, 0.1], E=[None]*3)
DT = dt.Frame(**DT0, **DT1)
assert_equals(DT, dt.cbind(DT0, DT1))
def test_cbind_expanded_frame():
DT = dt.Frame(A=[1, 2], B=['a', "E"], C=[7, 1000], D=[-3.14, 159265])
RES = dt.cbind(*DT)
assert_equals(DT, RES)
