python-chemfp 1.1p1-2.1 / usr / lib / python2.7 / dist-packages / chemfp / commandline / simsearch.py

from __future__ import with_statement
import math
import sys
import itertools
import time

import chemfp
from chemfp import argparse, io, SOFTWARE, bitops
from chemfp import search

# Suppose you have a 4K fingerprint.
#   1/4096 = 0.000244140625.
#   2/4096 = 0.00048828125
# You only need to show "0.0002" and "0.0005" to
# disambiguate the scores. I don't like seeing only
# the minimum resolution, so I also show at least
# the next bit.
#   For 4096 the float_formatter is %.5f and the
# above values are 0.00024 and 0.00049.
# This also prevents the results from being shown
# in scientific notation.
def get_float_formatter(num_bytes):
    num_digits = int(math.log10(num_bytes*8)) + 2
    float_formatter = "%." + str(num_digits) + "f"
    return float_formatter

def write_simsearch_magic(outfile):
    outfile.write("#Simsearch/1\n")

def write_count_magic(outfile):
    outfile.write("#Count/1\n")

def write_simsearch_header(outfile, d):
    lines = []
    for name in ("num_bits", "type", "software", "queries", "targets"):
        value = d.get(name, None)
        if value is not None:
            lines.append("#%s=%s\n" % (name, value))
    for name in ("query_sources", "target_sources"):
        for value in d.get(name, []):
            lines.append("#%s=%s\n" % (name, value))
    outfile.writelines(lines)

#### The NxM cases

def report_threshold(outfile, query_arenas, targets, threshold):
    float_formatter = get_float_formatter(targets.metadata.num_bytes)
    def search_function(query_arena):
        return targets.threshold_tanimoto_search_arena(query_arena, threshold=threshold)
    _report_search(outfile, float_formatter, query_arenas, search_function)

def report_knearest(outfile, query_arenas, targets, k, threshold):
    float_formatter = get_float_formatter(targets.metadata.num_bytes)
    def search_function(query_arena):
        return targets.knearest_tanimoto_search_arena(query_arena, k=k, threshold=threshold)
    _report_search(outfile, float_formatter, query_arenas, search_function)

def _report_search(outfile, float_formatter, query_arenas, search_function):
    hit_formatter = "\t%s\t" + float_formatter
    for query_arena in query_arenas:
        for query_id, row in zip(query_arena.ids, search_function(query_arena)):
            outfile.write("%d\t%s" % (len(row), query_id))
            for hit in row.get_ids_and_scores():
                outfile.write(hit_formatter % hit)
            outfile.write("\n") # XXX flush?
    


def report_counts(outfile, query_arenas, targets, threshold):
    for query_arena in query_arenas:
        counts = targets.count_tanimoto_hits_arena(query_arena, threshold)
        for query_id, hit_count in zip(query_arena.ids, counts):
            outfile.write("%d\t%s\n" % (hit_count, query_id))

#### The NxN cases


def do_NxN_searches(args, k, threshold, target_filename):
    t1 = time.time()

    # load_fingerprints sorts the fingerprints based on popcount
    # I want the output to be in the same order as the input.
    # This means I need to do some reordering. Consider:
    #   0003  ID_A
    #   010a  ID_B
    #   1000  ID_C
    # I use this to generate:
    #   original_ids = ["ID_A", "ID_B", "ID_C"]
    #   targets.ids = [2, 0, 1]
    #   original_index_to_current_index = {2:0, 0:1, 1:2}
    #   current_index_to_original_index = {0:2, 1:0, 2:1}
    
    original_ids = []
    fps = chemfp.open(target_filename)
    def get_index_to_id(fps):
        for i, (id, fp) in enumerate(fps):
            original_ids.append(id)
            yield i, fp
    
    targets = chemfp.load_fingerprints(get_index_to_id(fps), fps.metadata)
    original_index_to_current_index = dict(zip(targets.ids, xrange(len(targets))))
    current_index_to_original_id = dict((i, original_ids[original_index])
                                            for i, original_index in enumerate(targets.ids))
    
    t2 = time.time()
    outfile = io.open_output(args.output)
    with io.ignore_pipe_errors:
        type = "Tanimoto k=%(k)s threshold=%(threshold)s NxN=full" % dict(
            k=k, threshold=threshold, max_score=1.0)

        if args.count:
            type = "Count threshold=%(threshold)s NxN=full" % dict(
                threshold=threshold)
            write_count_magic(outfile)
        else:
            write_simsearch_magic(outfile)

        write_simsearch_header(outfile, {
            "num_bits": targets.metadata.num_bits,
            "software": SOFTWARE,
            "type": type,
            "targets": target_filename,
            "target_sources": targets.metadata.sources})

        if args.count:
            counts = search.count_tanimoto_hits_symmetric(targets, threshold,
                                                          batch_size=args.batch_size)
            for original_index, original_id in enumerate(original_ids):
                current_index = original_index_to_current_index[original_index]
                count = counts[current_index]
                outfile.write("%d\t%s\n" % (count, original_id))
        else:
            hit_formatter = "\t%s\t" + get_float_formatter(targets.metadata.num_bytes)
            if k == "all":
                results = search.threshold_tanimoto_search_symmetric(targets, threshold,
                                                                     batch_size=args.batch_size)
            else:
                results = search.knearest_tanimoto_search_symmetric(targets, k, threshold,
                                                                    batch_size=args.batch_size)

            for original_index, original_id in enumerate(original_ids):
                current_index = original_index_to_current_index[original_index]
                new_indices_and_scores = results[current_index].get_ids_and_scores()
                outfile.write("%d\t%s" % (len(new_indices_and_scores), original_id))
                for (new_index, score) in new_indices_and_scores:
                    original_id = original_ids[new_index]
                    outfile.write(hit_formatter % (original_id, score))
                outfile.write("\n") # XXX flush?

    t3 = time.time()
    if args.times:
        sys.stderr.write("open %.2f search %.2f total %.2f\n" % (t2-t1, t3-t2, t3-t1))

####

def int_or_all(s):
    if s == "all":
        return s
    return int(s)

# the "2fps" options need a way to say "get the options from --reference"
# ob2fps --reference targets.fps | simsearch -k  5 --threshold 0.5 targets.fps


parser = argparse.ArgumentParser(
    description="Search an FPS file for similar fingerprints")
parser.add_argument("-k" ,"--k-nearest", help="select the k nearest neighbors (use 'all' for all neighbors)",
                    default=None, type=int_or_all)
parser.add_argument("-t" ,"--threshold", help="minimum similarity score threshold",
                    default=None, type=float)
parser.add_argument("-q", "--queries", help="filename containing the query fingerprints")
parser.add_argument("--NxN", action="store_true",
                    help="use the targets as the queries, and exclude the self-similarity term")
parser.add_argument("--hex-query", help="query in hex")
parser.add_argument("--query-id", default="Query1",
                    help="id for the hex query")
parser.add_argument("--in", metavar="FORMAT", dest="query_format",
                    help="input query format (default uses the file extension, else 'fps')")
parser.add_argument("-o", "--output", metavar="FILENAME",
                    help="output filename (default is stdout)")

parser.add_argument("-c", "--count", help="report counts", action="store_true")

parser.add_argument("-b", "--batch-size", help="batch size",
                    default=100, type=int)

parser.add_argument("--scan", help="scan the file to find matches (low memory overhead)",
                    action="store_true")
parser.add_argument("--memory", help="build and search an in-memory data structure (faster for multiple queries)",
                    action="store_true")

parser.add_argument("--times", help="report load and execution times to stderr",
                    action="store_true")

parser.add_argument("target_filename", nargs=1, help="target filename", default=None)

## Something to enable multi-threading
#parser.add_argument("-j", "--jobs", help="number of jobs ",
#                    default=10, type=int)


def main(args=None):
    args = parser.parse_args(args)
    target_filename = args.target_filename[0]

    threshold = args.threshold
    k = args.k_nearest

    if args.count and k is not None and k != "all":
        parser.error("--count search does not support --k-nearest")

    # People should not use this without setting parameters.  On the
    # other hand, I don't want an error message if there are no
    # parameters. This solution seems to make sense.

    if threshold is None:
        if k is None:
            # If nothing is set, use defaults of --thresdhold 0.7 -k 3            
            threshold = 0.7
            k = 3
        else:
            # only k is set; search over all possible matches
            threshold = 0.0
    else:
        if k is None:
            # only threshold is set; search for all hits above that threshold
            k = "all"

    if k == "all":
        pass
    elif k < 0:
        parser.error("--k-nearest must be non-negative or 'all'")

    if not (0.0 <= threshold <= 1.0):
        parser.error("--threshold must be between 0.0 and 1.0, inclusive")

    if args.batch_size < 1:
        parser.error("--batch-size must be positive")

    bitops.use_environment_variables()

    if args.NxN:
        if args.scan:
            parser.error("Cannot specify --scan with an --NxN search")
        if args.hex_query:
            parser.error("Cannot specify --hex-query with an --NxN search")
        if args.queries:
            parser.error("Cannot specify --queries with an --NxN search")
        do_NxN_searches(args, k, threshold, target_filename)
        return
            
    if args.scan and args.memory:
        parser.error("Cannot specify both --scan and --memory")

    if args.hex_query and args.queries:
        parser.error("Cannot specify both --hex-query and --queries")
    if args.hex_query:
        query_id = args.query_id
        for c, name in ( ("\t", "tab"),
                         ("\n", "newline"),
                         ("\r", "control-return"),
                         ("\0", "NUL")):
            if c in query_id:
                parser.error("--query-id must not contain the %s character" %
                             (name,))
    

    # Open the target file. This reads just enough to get the header.

    try:
        targets = chemfp.open(target_filename)
    except (IOError, ValueError), err:
        sys.stderr.write("Cannot open targets file: %s" % err)
        raise SystemExit(1)

    if args.hex_query is not None:
        try:
            query_fp = args.hex_query.decode("hex")
        except TypeError, err:
            parser.error("--hex-query is not a hex string: %s" % (err,))

        for (severity, error, msg_template) in chemfp.check_fp_problems(query_fp, targets.metadata):
            if severity == "error":
                parser.error(msg_template % dict(fp="query", metadata=repr(target_filename)))
            
        num_bits = targets.metadata.num_bits
        if num_bits is None:
            num_bits = len(query_fp) * 8
        query_metadata = chemfp.Metadata(num_bits=num_bits, num_bytes=len(query_fp))
        queries = chemfp.Fingerprints(query_metadata,
                                      [(query_id, query_fp)])
        query_filename = None
    else:
        query_filename = args.queries
        try:
            queries = chemfp.open(query_filename, format=args.query_format)
        except (ValueError, IOError), err:
            sys.stderr.write("Cannot open queries file: %s\n" % (err,))
            raise SystemExit(1)

    batch_size = args.batch_size
    query_arena_iter = queries.iter_arenas(batch_size)
    
    t1 = time.time()

    first_query_arena = None
    for first_query_arena in query_arena_iter:
        break

    if args.scan:
        # Leave the targets as-is
        pass
    elif args.memory:
        targets = chemfp.load_fingerprints(targets)
    if not first_query_arena:
        # No input. Leave as-is
        pass
    elif len(first_query_arena) < min(10, batch_size):
        # Figure out the optimal search. If there is a
        # small number of inputs (< ~10) then a scan
        # of the FPS file is faster than an arena search.
        pass
    else:
        targets = chemfp.load_fingerprints(targets)

    problems = chemfp.check_metadata_problems(queries.metadata, targets.metadata)
    for (severity, error, msg_template) in problems:
        msg = msg_template % dict(metadata1="queries", metadata2="targets")
        if severity == "error":
            parser.error(msg)
        elif severity == "warning":
            sys.stderr.write("WARNING: " + msg + "\n")

    t2 = time.time()
    outfile = io.open_output(args.output)
    with io.ignore_pipe_errors:
        type = "Tanimoto k=%(k)s threshold=%(threshold)s" % dict(
            k=k, threshold=threshold, max_score=1.0)

        if args.count:
            type = "Count threshold=%(threshold)s" % dict(
                threshold=threshold)
            write_count_magic(outfile)
        else:
            write_simsearch_magic(outfile)

        write_simsearch_header(outfile, {
            "num_bits": targets.metadata.num_bits,
            "software": SOFTWARE,
            "type": type,
            "queries": query_filename,
            "targets": target_filename,
            "query_sources": queries.metadata.sources,
            "target_sources": targets.metadata.sources})

        if first_query_arena:
            query_arenas = itertools.chain([first_query_arena],
                                           query_arena_iter)

            if args.count:
                report_counts(outfile, query_arenas, targets,
                              threshold = threshold)
            elif k == "all":
                report_threshold(outfile, query_arenas, targets,
                                 threshold = threshold)
            else:
                report_knearest(outfile, query_arenas, targets,
                                k = k, threshold = threshold)
                                
                
                    
    t3 = time.time()
    if args.times:
        sys.stderr.write("open %.2f search %.2f total %.2f\n" % (t2-t1, t3-t2, t3-t1))

if __name__ == "__main__":
    main()