#!/usr/bin/env python

import getopt
import sys
import tempfile
import string

import sys
sys.path.append("liblinear-1.8/python")
import liblinearutil as ll

import common
import parse
import vectorize

def usage(f = sys.stdout):
    print >> f, """\
Usage: %s -m <MODEL_FILENAME>
Saves a liblinear model in C struct format.

  -h, --help                  show this help.
  -m, --model=MODEL_FILENAME  use the model in MODEL_FILENAME (required).\
""" % sys.argv[0]

class options (object):
    model_filename = None

def c_quote(s):
    r = []
    for c in s:
        if c == "\"":
            c = "\\\""
        elif c == "\\":
            c = "\\\\"
        elif c not in string.printable:
            c = "\\" + "%03o" % ord(c)
        r.append(c)
    return '"' + "".join(r) + '"'

def c_quote_or_null(s):
    if s is None:
        return "NULL";
    else:
        return c_quote(s)

def format_osclass(osclass):
    assert len(osclass) == 4, osclass

    return "{" + ", ".join(c_quote_or_null(x) for x in osclass) + "}"

def save_model_c_struct(f, model):
    """Write a liblinear model as a C struct."""
    if model.model.bias >= 0:
        w_size = model.model.nr_feature + 1
    else:
        w_size = model.model.nr_feature
    if model.model.nr_class == 2 and model.model.solver_type != 4:
        nr_w = 1
    else:
        nr_w = model.model.nr_class

    f.write("/* This file is automatically generated. */\n")

    f.write("\n")

    f.write("#include \"FingerPrintResults.h\"\n")
    f.write("#include \"linear.h\"\n")

    f.write("\n")

    f.write("""\
/* Scale parameters are pairs (a, b). A scaled value x' is calculated from
   a, b, and an observed x by x' = (x + a) * b. */
""")
    f.write("double FPscale[][2] = {\n")
    if model.scale_params:
        for i in range(w_size):
            mn, mx = model.scale_params[i]
            a = -mn
            if mn == mx:
                b = 1.0
            else:
                b = 1.0 / (mx - mn)
            f.write("\t{ %10g, %10g },\t/* %s */\n" % (a, b, model.feature_names[i]))
    else:
        for i in range(w_size):
            f.write("\t{ 0.0, 1.0 },\n")
    f.write("};\n")

    f.write("\n")

    f.write("double FPmean[][%d] = {\n" % w_size)
    for desc, means in zip(model.descs, model.means):
        f.write("\t/* %s */\n" % desc.nmapname);
        f.write("\t{ ")
        assert len(means) == w_size, (len(means), nr_w)
        for m in means:
            f.write("%+.8f, " % m);
        f.write("},\n")
    f.write("};\n")

    f.write("\n")

    f.write("""\
/* Variances are per-feature only, not considering correlation. Rows of this
   array can be thought of as diagonals of a covariance matrices, under the
   assumption that feature values are independent (which is not true in
   general). This is an approximation to avoid having to store an n^2 covariance
   matrix per OS class. */
""")
    f.write("double FPvariance[][%d] = {\n" % w_size)
    for desc, variances in zip(model.descs, model.variances):
        f.write("\t/* %s */\n" % desc.nmapname);
        f.write("\t{ ")
        assert len(variances) == w_size, (len(variances), nr_w)
        for v in variances:
            f.write("%+.8f, " % v);
        f.write("},\n")
    f.write("};\n")

    f.write("\n")

    f.write("/* Regression coefficients. Columns are classes, rows are features. */\n")
    f.write("static double _w[] = {\n");
    for i in range(w_size):
        f.write("\t/* %s */\n" % model.feature_names[i])
        f.write("\t")
        for j in range(nr_w):
            f.write("%+.8f, " % model.model.w[i * nr_w + j])
        f.write("\n")
    f.write("};\n")

    f.write("\n")

    f.write("static int _labels[] = {" + ", ".join(str(x) for x in range(model.model.nr_class)) + "};\n")

    f.write("\n")

    f.write("/* C = %f */\n" % model.model.param.C)
    f.write("struct model FPModel = {\n")
    f.write("\t{%d},\n" % model.model.param.solver_type)
    f.write("\t%d,\n" % model.model.nr_class)
    f.write("\t%d,\n" % model.model.nr_feature)
    f.write("\t_w,\n");
    f.write("\t_labels,\n");
    f.write("\t%.8f\n" % model.model.bias)
    f.write("};\n")

    f.write("\n");

    f.write("std::vector<FingerMatch> load_fp_matches() {\n")
    f.write("\tstd::vector<FingerMatch> matches;\n");
    for i, desc in enumerate(model.descs):
        f.write("\t{\n");
        f.write("\t\tFingerMatch match;\n");
        f.write("\t\tmatch.line = %d;\n" % i);
        f.write("\t\tmatch.OS_name = (char *) %s;\n" % c_quote(desc.nmapname));
        for nmapclass in desc.nmapclasses:
            f.write("\t\t{\n")
            f.write("\t\t\tOS_Classification osclass = %s;\n" % format_osclass(nmapclass.nmapclass))
            for cpe in nmapclass.cpe:
                f.write("\t\t\tosclass.cpe.push_back(%s);\n" % c_quote(cpe))
            f.write("\t\t\tmatch.OS_class.push_back(osclass);\n")
            f.write("\t\t}\n")
        f.write("\t\tmatches.push_back(match);\n");
        f.write("\t}\n");
    f.write("\treturn matches;\n");
    f.write("}\n");

opts, args = getopt.gnu_getopt(sys.argv[1:], "hm:", ["help", "model="])
for o, a in opts:
    if o == "-h" or o == "--help":
        usage()
        sys.exit()
    elif o == "-m" or o == "--model":
        options.model_filename = a

if options.model_filename is None or args:
    usage(sys.stderr)
    exit(1)

model = parse.parse_model_file(options.model_filename)

save_model_c_struct(sys.stdout, model)