Disable scientific notation in python json.dumps output
One way to format
evil = {"x": 0.00000000001}
is to steal Decimal
's "f" formatter. It's the only easy way I've found that avoids both cropping problems and exponents, but it's not space efficient.
class FancyFloat(float): def __repr__(self): return format(Decimal(self), "f")
To use it you can make an encoder that "decimalize"s the input
class JsonRpcEncoder(json.JSONEncoder): def decimalize(self, val): if isinstance(val, dict): return {k:self.decimalize(v) for k,v in val.items()} if isinstance(val, (list, tuple)): return type(val)(self.decimalize(v) for v in val) if isinstance(val, float): return FancyFloat(val) return val def encode(self, val): return super().encode(self.decimalize(val))JsonRpcEncoder().encode(evil)#>>> '{"x": 0.00000000000999999999999999939496969281939810930172340963650867706746794283390045166015625}'
or, of course, you could move the decimalization out into a function and call that before json.dumps
.
That's how I would do it, even if it's a lame method.
I can't find an answer to avoid the problem that converts 0.00001
to 1e-5
,so I wrote a pretty_float_json_dumps
function. It works fine in my project! Hope it can help someone!!
def pretty_float_json_dumps(json_obj): dumps_str = "" if isinstance(json_obj, dict): dumps_str += "{" for k,v in json_obj.items(): dumps_str += json.dumps(k)+":" if isinstance(v, float): float_tmp_str = ("%.16f" % v).rstrip("0") dumps_str += (float_tmp_str+'0' if float_tmp_str.endswith('.') else float_tmp_str) + ',' elif isinstance(v, list) or isinstance(v, dict): dumps_str += pretty_float_json_dumps(v)+',' else: dumps_str += pretty_float_json_dumps(v)+',' if dumps_str.endswith(','): dumps_str = dumps_str[:-1] dumps_str += "}" elif isinstance(json_obj, list): dumps_str += "[" for v in json_obj: if isinstance(v, float): float_tmp_str = ("%.16f" % v).rstrip("0") dumps_str += (float_tmp_str+'0' if float_tmp_str.endswith('.') else float_tmp_str) + ',' elif isinstance(v, list) or isinstance(v, dict): dumps_str += pretty_float_json_dumps(v)+',' else: dumps_str += pretty_float_json_dumps(v)+',' if dumps_str.endswith(','): dumps_str = dumps_str[:-1] dumps_str += "]" else: dumps_str += json.dumps(json_obj) return dumps_str
This is supplemental commentary, not a complete answer for avoiding scientific notation in json.dumps
. With an extra round of parsing, the parse_float
option on json.loads
can help with some things, e.g.
$ pythonPython 3.7.10 | packaged by conda-forge | (default, Feb 19 2021, 16:07:37) [GCC 9.3.0] on linuxType "help", "copyright", "credits" or "license" for more information.>>> import json>>> data = {"x": 0.0000001}>>> json.dumps(data)'{"x": 1e-07}'>>> data = json.loads(json.dumps(data), parse_float=lambda x: round(float(x), 6))>>> json.dumps(data)'{"x": 0.0}'
This only avoids scientific notation for small values that are rounded to zero. While that is not efficient for large datasets, it can help in some use-cases. This does not completely avoid scientific notation.
>>> data = {"x": 0.00001}>>> data = json.loads(json.dumps(data), parse_float=lambda x: round(float(x), 6))>>> json.dumps(data)'{"x": 1e-05}'