Consistent CategoricalDtype use in Categorical init · pandas-dev/pandas@416d1d7 (original) (raw)

`@@ -234,6 +234,21 @@ class Categorical(PandasObject):

234

`def init(self, values, categories=None, ordered=None, dtype=None,

235

`fastpath=False):

236

237

Ways of specifying the dtype (prioritized ordered)

238

1. dtype is a CategoricalDtype

239

a.) with known categories, use dtype.categories

240

b.) else with Categorical values, use values.dtype

241

c.) else, infer from values

242

d.) specifying dtype=CategoricalDtype and categories is an error

243

2. dtype is a string 'category'

244

a.) use categories, ordered

245

b.) use values.dtype

246

c.) infer from values

247

3. dtype is None

248

a.) use categories, ordered

249

b.) use values.dtype

250

c.) infer from values

251

+

237

252

`if dtype is not None:

238

253

`if isinstance(dtype, compat.string_types):

239

254

`if dtype == 'category':

`@@ -247,20 +262,24 @@ def init(self, values, categories=None, ordered=None, dtype=None,

247

262

`categories = dtype.categories

248

263

`ordered = dtype.ordered

249

264

250

if ordered is None:

251

ordered = False

265

elif is_categorical(values):

266

dtype = values.dtype._from_categorical_dtype(values.dtype,

267

categories, ordered)

268

else:

269

dtype = CategoricalDtype(categories, ordered)

270

+

271

At this point, dtype is always a CategoricalDtype

272

if dtype.categories is None, we are inferring

252

273

253

274

`if fastpath:

254

if dtype is None:

255

dtype = CategoricalDtype(categories, ordered)

256

275

`self._codes = coerce_indexer_dtype(values, categories)

257

276

`self._dtype = dtype

258

277

`return

259

278

260

279

`# sanitize input

261

280

`if is_categorical_dtype(values):

262

281

263

we are either a Series, CategoricalIndex

282

we are either a Series or a CategoricalIndex

264

283

`if isinstance(values, (ABCSeries, ABCCategoricalIndex)):

265

284

`values = values._values

266

285

`@@ -271,6 +290,7 @@ def init(self, values, categories=None, ordered=None, dtype=None,

271

290

`values = values.get_values()

272

291

273

292

`elif isinstance(values, (ABCIndexClass, ABCSeries)):

293

we'll do inference later

274

294

`pass

275

295

276

296

`else:

`@@ -288,12 +308,12 @@ def init(self, values, categories=None, ordered=None, dtype=None,

288

308

`# "object" dtype to prevent this. In the end objects will be

289

309

`# casted to int/... in the category assignment step.

290

310

`if len(values) == 0 or isna(values).any():

291

dtype = 'object'

311

sanitize_dtype = 'object'

292

312

`else:

293

dtype = None

294

values = _sanitize_array(values, None, dtype=dtype)

313

sanitize_dtype = None

314

values = _sanitize_array(values, None, dtype=sanitize_dtype)

295

315

296

if categories is None:

316

if dtype.categories is None:

297

317

`try:

298

318

`codes, categories = factorize(values, sort=True)

299

319

`except TypeError:

`@@ -310,7 +330,8 @@ def init(self, values, categories=None, ordered=None, dtype=None,

310

330

`raise NotImplementedError("> 1 ndim Categorical are not "

311

331

`"supported at this time")

312

332

313

if dtype is None or isinstance(dtype, str):

333

if dtype.categories is None:

334

we're inferring from values

314

335

`dtype = CategoricalDtype(categories, ordered)

315

336

316

337

`else:

`@@ -321,11 +342,6 @@ def init(self, values, categories=None, ordered=None, dtype=None,

321

342

`# - the new one, where each value is also in the categories array

322

343

`# (or np.nan)

323

344

324

make sure that we always have the same type here, no matter what

325

we get passed in

326

if dtype is None or isinstance(dtype, str):

327

dtype = CategoricalDtype(categories, ordered)

328

-

329

345

`codes = _get_codes_for_values(values, dtype.categories)

330

346

331

347

`# TODO: check for old style usage. These warnings should be removes