class DaskCollect(OpVertex):
"""
Calls dask.compute on a collection of dask delayed objects
"""
dask_progress_bar = BaseParameter(
True, "Include a diagnostic Progressbar from dask")
num_workers = BaseParameter(4, "Number of dask workers")
scheduler = BaseParameter('processes', "Dask scheduler option")
def _node_color(self):
return '#2fbc2d'
def run(self, *ops):
self.ops = list()
self.delays = list()
for _op, _delay in ops:
self.ops.append(_op)
self.delays.append(_delay)
compute_kwargs = dict(scheduler=self.scheduler,
num_workers=self.num_workers)
if self.dask_progress_bar:
with ProgressBar():
results = dask.compute(*self.delays, **compute_kwargs)
else:
results = dask.compute(*self.delays, **compute_kwargs)
return results
class DaskParallel(OpVertex):
"""
Wraps an Op in a dask.delayed object
"""
parallel_op = BaseParameter()
op_kwargs = BaseParameter(dict())
def _node_color(self):
return '#2fbc2d'
def _node_shape(self):
#return 'doublecircle'
return 'doubleoctagon'
def requires(self):
if inspect.isclass(self.parallel_op):
# instantiate with provided keyword args
self.inst_op = self.parallel_op(**self.op_kwargs)
else:
# already instantiated
self.inst_op = self.parallel_op
return self.inst_op.requires()
def run(self, *args, **kwargs):
return (self.inst_op, dask.delayed(self.inst_op.run)(*args, **kwargs))
def get_name(self):
return self.parallel_op.get_name()
class DropDuplicates(DFOp):
subset = BaseParameter(None)
keep = BaseParameter('first')
inplace = BaseParameter(False)
def run(self, df):
return df.drop_duplicates(subset=self.subset,
keep=self.keep,
inplace=self.inplace)
class RenameColumns(DFOp):
columns = BaseParameter(None)
copy = BaseParameter(True)
inplace = BaseParameter(False)
level = BaseParameter(None)
def run(self, df):
return df.rename(columns=self.columns,
copy=self.copy,
inplace=self.inplace,
level=self.level)
class DropNa(DFOp):
axis = BaseParameter(0)
how = BaseParameter('any')
thresh = BaseParameter(None)
subset = BaseParameter(None)
def run(self, df):
return df.dropna(axis=self.axis,
how=self.how,
thresh=self.thresh,
subset=self.subset,
inplace=False)
class AssignColumn(DFOp):
column = BaseParameter(None)
value = BaseParameter(None)
assignments = BaseParameter(None)
def run(self, df):
if self.assignments is not None:
for col, val in self.assignments.items():
df[col] = val
else:
df[self.column] = self.value
return df
class Read_CSV(OpVertex):
filepath_or_buffer = BaseParameter()
sep = BaseParameter(',')
delimiter = BaseParameter(None)
header = BaseParameter('infer')
def run(self, path=None):
if path is None:
path = self.filepath_or_buffer
return pd.read_csv(filepath_or_buffer=path,
sep=self.sep,
delimiter=self.delimiter,
header=self.header)
class HistMulti(OpVertex):
title = BaseParameter('')
figsize = BaseParameter((8, 5))
xlabel = BaseParameter('x')
ylabel = BaseParameter('y')
logx = BaseParameter(False)
logy = BaseParameter(False)
passthrough = BaseParameter(False,
"If False, returns axis of plot rather ")
# @staticmethod
def hist_compare(self, **hist_data):
fig, ax = plt.subplots(figsize=self.figsize)
hist_kwargs = dict(bins=20, alpha=.4, density=False)
for d_name, d in hist_data.items():
ax.hist(d, label=d_name, **hist_kwargs)
ax.legend(fontsize=15)
ax.set_title(self.title, fontsize=18)
ax.set_xlabel(self.xlabel, fontsize=16)
ax.set_ylabel(self.ylabel, fontsize=16)
# ax.set_xticklabels(ax.get_xticklabels(), fontsize=13)
ax.tick_params(labelsize=15)
if self.logy:
ax.set_yscale('log')
if self.logx:
ax.set_xscale('log')
if self.passthrough:
return hist_data
else:
return ax
class Subplots(OpVertex):
nrows = BaseParameter(1)
ncols = BaseParameter(1)
sharex = BaseParameter(False)
sharey = BaseParameter(False)
#squeeze=BaseParameter(True)
subplot_kw = BaseParameter(None)
gridspec_kw = BaseParameter(None)
figsize = BaseParameter((12, 6))
_never_cache = True
#fig_kw = dsdag.core.parameter.BaseParameter(dict())
def run(self):
from matplotlib import pyplot as plt
#self.fig, self.axs = plt.subplots(**{k: v.value for k, v in self.get_parameters()})
self.fig, self.axs = plt.subplots(nrows=self.nrows,
ncols=self.ncols,
sharex=self.sharex,
sharey=self.sharey,
figsize=self.figsize,
subplot_kw=self.subplot_kw,
gridspec_kw=self.gridspec_kw,
squeeze=False)
self.axs = [
self.axs[r][c] for r in range(self.nrows)
for c in range(self.ncols)
]
return self.axs
class Join(DFOp):
how = BaseParameter('left')
lsuffix = BaseParameter('')
rsuffix = BaseParameter('')
sort = BaseParameter(False)
def requires(self):
raise NotImplementedError()
def run(self, *args):
assert all(isinstance(o, pd.DataFrame) for o in args)
ret = args[0]
for o in args[1:]:
ret = ret.join(o,
how=self.how,
lsuffix=self.lsuffix,
rsuffix=self.rsuffix,
sort=self.sort)
return ret
class LambdaOp(OpVertex):
f = BaseParameter(help_msg="Function that is applied to the input")
def _node_color(self):
return '#d65768'
def requires(self):
raise NotImplementedError("Incomplete LambdaOp - must be applied")
def run(self, *args, **kwargs):
return self.f(*args, **kwargs)
class FrameBrowse(DFOp, FrameBrowseMaixin):
passthrough = BaseParameter(True)
def op_nb_viz(self, op_out, viz_out=None):
return FrameBrowseMaixin.op_nb_viz(op_out, viz_out)
def run(self, df):
output = self.op_nb_viz(df)
display(output)
if self.passthrough:
return df
class Concat(DFOp):
axis = BaseParameter(0)
join = BaseParameter('outer')
ignore_index = BaseParameter(False)
keys = BaseParameter(None)
levels = BaseParameter(None)
verify_integrity = BaseParameter(False)
sort = BaseParameter(None)
copy = BaseParameter(True)
def requires(self):
raise NotImplementedError()
def run(self, *args):
return pd.concat(args,
axis=self.axis,
join=self.join,
ignore_index=self.ignore_index,
keys=self.keys,
levels=self.levels,
verify_integrity=self.verify_integrity,
sort=self.sort,
copy=self.copy)
class ContourPlot(OpVertex):
sample_ixes = BaseParameter(None)
title = BaseParameter('')
bands = BaseParameter(None)
value_column = BaseParameter(None)
dot_color_column = BaseParameter(None)
cbar = BaseParameter(False)
_never_cache = True
@staticmethod
def contour_plot(_s_df,
value_column,
dot_color=None,
vmin=None,
vmax=None,
ax=None,
cbar=False,
title=''):
if ax is None:
fig, ax = plt.subplots()
x = _s_df['x']
y = _s_df['y']
z = _s_df[value_column]
# define grid.
_s_2d_df = _s_df.dropna().pivot(index='x',
columns='y',
values=value_column)
_s_stacked_df = _s_2d_df.stack(
dropna=False).rename(value_column).reset_index()
# return _s_2d_df
xi = _s_2d_df.index.tolist()
yi = _s_2d_df.columns.tolist()
zi = griddata(x, y, z, xi, yi, interp='linear')
vmax = abs(zi).max() if vmax is None else vmax
vmin = -abs(zi).max() if vmin is None else vmin
ctr = ax.contourf(xi,
yi,
zi,
30,
cmap=plt.cm.hsv,
vmax=vmax,
vmin=vmin)
if cbar:
#plt.colorbar(mappable=ctr)
plt.colorbar(ctr, ax=ax)
ax.scatter(
x=_s_df['x'],
y=_s_df['y'],
cmap='gray',
s=5,
c=dot_color,
)
ax.set_title(title)
return ax
def run(
self,
df,
ax=None,
value_column=None,
):
value_column = self.value_column if value_column is None else value_column
plt_df = df.dropna()
self.contour_plot(
plt_df,
value_column=value_column,
title=self.title,
ax=ax,
dot_color=plt_df[value_column] if self.dot_color_column is None
else plt_df[self.dot_color_column],
cbar=self.cbar)
class DenseAutoencoder(OpVertex):
input_feature_dim = BaseParameter()
output_feature_dim = BaseParameter()
encoding_dim = BaseParameter(25)
dropout = BaseParameter(.15)
input_dropout = BaseParameter(0.15)
width = BaseParameter(128)
depth = BaseParameter(1)
hidden_activation = BaseParameter('relu')
output_activation = BaseParameter('relu')
lr = BaseParameter(0.001)
decay = BaseParameter(0.0)
optimizer = BaseParameter(None)
loss = BaseParameter('mean_squared_error')
batchnorm = BaseParameter(True)
kernel_initializer = BaseParameter('glorot_normal')
bias_initializer = BaseParameter('zeros')
@staticmethod
def build_autoencoder(input_feature_dim,
output_feature_dim,
encoding_dim=25,
dropout=.15,
input_dropout=0.15,
width=128,
depth=1,
hidden_activation='relu',
output_activation='relu',
lr=0.001,
decay=0.0,
optimizer=None,
loss='mean_squared_error',
batchnorm=True,
kernel_initializer='glorot_normal',
bias_initializer='zeros'):
"""
returns encoder, decoder, autoencoder
"""
#print("Input-Enc-Output: %d-%d-%d" % (input_feature_dim, encoding_dim, output_feature_dim))
input_sym = Input(shape=(input_feature_dim, ))
enc_x = input_sym
if input_dropout is not None and input_dropout > 0.:
enc_x = Dropout(input_dropout)(enc_x)
for d in range(depth):
enc_x = Dense(width,
activation=None,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer)(enc_x)
if batchnorm:
enc_x = BatchNormalization()(enc_x)
enc_x = Activation(hidden_activation)(enc_x)
enc_x = Dropout(dropout)(enc_x)
encoded = Dense(encoding_dim,
activation=None,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer)(enc_x)
if batchnorm:
encoded = BatchNormalization()(encoded)
encoded = Activation(hidden_activation, name='encoder')(encoded)
decoder = Dropout(dropout)(encoded)
dec_x = decoder
for d in range(depth):
dec_x = Dense(width,
activation=None,
name='decoder_l%d' % d,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer)(dec_x)
if batchnorm:
dec_x = BatchNormalization()(dec_x)
dec_x = Activation(hidden_activation)(dec_x)
dec_x = Dropout(dropout)(dec_x)
decoded = Dense(output_feature_dim,
activation=None,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer)(dec_x)
decoded = Activation(output_activation)(decoded)
autoencoder = Model(input_sym, decoded)
encoder = Model(input_sym, encoded)
# create a placeholder for an encoded (32-dimensional) input
encoded_input = Input(shape=(encoding_dim, ))
# retrieve the last layer of the autoencoder model
# decoder_layer = autoencoder.layers[-(depth*4 + 1)]
decoder_layer = [
l for l in autoencoder.layers if l.name == 'decoder_l0'
][0]
# create the decoder model
decoder = Model(encoded_input, decoder_layer(encoded_input))
if optimizer is None:
optimizer = Adam(lr=lr, decay=decay)
autoencoder.compile(optimizer=optimizer, loss=loss)
return encoder, decoder, autoencoder
class ApplyMap(DFOp):
func = BaseParameter()
def run(self, df):
return df.applymap(func=self.func)
class Query(DFOp):
q = BaseParameter()
def run(self, df):
return df.query(self.q)
class SelectColumns(DFOp):
columns = BaseParameter(None)
def run(self, df):
return df[self.columns]
class Merge(DFOp):
#key = BaseParameter()
how = BaseParameter('inner')
on = BaseParameter(None)
left_on = BaseParameter(None)
right_on = BaseParameter(None)
left_index = BaseParameter(False)
right_index = BaseParameter(False)
sort = BaseParameter(False)
suffixes = BaseParameter(('_x', '_y'))
copy = BaseParameter(True)
indicator = BaseParameter(False)
validate = BaseParameter(None)
def requires(self):
raise NotImplementedError()
def run(self, *args):
#frames = list(kwargs.values())
frames = list(args)
merged = frames[0]
for f in frames[1:]:
merged = merged.merge(f,
on=self.on,
how=self.how,
left_on=self.left_on,
right_on=self.right_on,
left_index=self.left_index,
right_index=self.right_index,
sort=self.sort,
suffixes=self.suffixes,
copy=self.copy,
indicator=self.indicator,
validate=self.validate)
return merged
class Select(OpVertex):
i = BaseParameter(0)
_never_cache = True
def run(self, l):
return l[self.i]
class InputOp(OpVertex):
obj = BaseParameter(help_msg="The object to wrap and return")
def run(self):
return self.obj
class Drop(DFOp):
labels = BaseParameter()
axis = BaseParameter(0)
def run(self, df):
return df.drop(labels=self.labels, axis=self.axis)
